598

I need help with this integral:

I=111x1+x1xln(2x2+2x+12x22x+1) dx.

The integrand graph looks like this:

The integrand graph

The approximate numeric value of the integral:

I8.372211626601275661625747121...

Neither Mathematica nor Maple could find a closed form for this integral, and lookups of the approximate numeric value in WolframAlpha and ISC+ did not return plausible closed form candidates either. But I still hope there might be a closed form for it.

I am also interested in cases when only numerator or only denominator is present under the logarithm.

CC BY-SA 4.0
10

14 Answers 14

Reset to default
1013
+150

I will transform the integral via a substitution, break it up into two pieces and recombine, perform an integration by parts, and perform another substitution to get an integral to which I know a closed form exists. From there, I use a method I know to attack the integral, but in an unusual way because of the 8th degree polynomial in the denominator of the integrand.

First sub t=(1x)/(1+x), dt=2/(1+x)2dxto get

20dtt1/21t2log(52t+t212t+5t2)

Now use the symmetry from the map t1/t. Break the integral up into two as follows:

201dtt1/21t2log(52t+t212t+5t2)+21dtt1/21t2log(52t+t212t+5t2)=201dtt1/21t2log(52t+t212t+5t2)+201dtt1/21t2log(52t+t212t+5t2)=201dtt1/21tlog(52t+t212t+5t2)

Sub t=u2to get

401du1u2log(52u2+u412u2+5u4)

Integrate by parts:

[2log(1+u1u)log(52u2+u412u2+5u4)]013201du(u56u3+u)(u42u2+5)(5u42u2+1)log(1+u1u)

One last sub: u=(v1)/(v+1)du=2/(v+1)2dv, and finally get

80dv(v21)(v46v2+1)v8+4v6+70v4+4v2+1logv

With this form, we may finally conclude that a closed form exists and apply the residue theorem to obtain it. To wit, consider the following contour integral:

Cdz8(z21)(z46z2+1)z8+4z6+70z4+4z2+1log2z

where Cis a keyhole contour about the positive real axis. This contour integral is equal to (I omit the steps where I show the integral vanishes about the circular arcs)

i4π0dv8(v21)(v46v2+1)v8+4v6+70v4+4v2+1logv+4π20dv8(v21)(v46v2+1)v8+4v6+70v4+4v2+1

It should be noted that the second integral vanishes; this may be easily seen by exploiting the symmetry about v1/v.

On the other hand, the contour integral is i2πtimes the sum of the residues about the poles of the integrand. In general, this requires us to find the zeroes of the eight degree polynomial, which may not be possible analytically. Here, on the other hand, we have many symmetries to exploit, e.g., if ais a root, then 1/ais a root, ais a root, and a¯is a root. For example, we may deduce that

z8+4z6+70z4+4z2+1=(z4+4z3+10z2+4z+1)(z44z3+10z24z+1)

which exploits the aasymmetry. Now write

z4+4z3+10z2+4z+1=(za)(za¯)(z1a)(z1a¯)

Write a=reiθand get the following equations:

(r+1r)cosθ=2
(r2+1r2)+4cos2θ=10

From these equations, one may deduce that a solution is r=ϕ+ϕand cosθ=1/ϕ, where ϕ=(1+5)/2is the golden ratio. Thus the poles take the form

zk=±(ϕ±ϕ)e±iarctanϕ

Now we have to find the residues of the integrand at these 8 poles. We can break this task up by computing:

k=18Resz=zk[8(z21)(z46z2+1)log2zz8+4z6+70z4+4z2+1]=k=18Resz=zk[8(z21)(z46z2+1)z8+4z6+70z4+4z2+1]log2zk

Here things got very messy, but the result is rather unbelievably simple:

Resz=zk[8(z21)(z46z2+1)z8+4z6+70z4+4z2+1]=sgn[cos(argzk)]

EDIT

Actually, this is a very simple computation. Inspired by @sos440, one may express the rational function of zin a very simple form:

8(z21)(z46z2+1)z8+4z6+70z4+4z2+1=[p(z)p(z)+p(z)p(z)]

where

p(z)=z4+4z3+10z2+4z+1

The residue of this function at the poles are then easily seen to be ±1according to whether the pole is a zero of p(z)or p(z).

END EDIT

That is, if the pole has a positive real part, the residue of the fraction is +1; if it has a negative real part, the residue is 1.

Now consider the log piece. Expanding the square, we get 3 terms:

log2|zk|(argzk)2+i2log|zk|argzk

Summing over the residues, we find that because of the ±1contributions above, that the first and third terms sum to zero. This leaves the second term. For this, it is crucial that we get the arguments right, as argzk[0,2π). Thus, we have

I=0dv8(v21)(v46v2+1)v8+4v6+70v4+4v2+1logv=12k=18sgn[cos(argzk)](argzk)2=12[2(arctanϕ)2+2(2πarctanϕ)22(πarctanϕ)22(π+arctanϕ)2]=2π24πarctanϕ=4πarccotϕ

CC BY-SA 3.0
12
  • 157
    +1. arccot(ϕ)-definitely one of the most weirdest closed form solutions to have ever been obtained!
    – user17762
    Commented Dec 2, 2013 at 6:31
  • 22
    @ShikariShambu: which makes me wonder if we can generate a list of the oddest closed-form solutions to integrals, sums, products, diff eq'ns, and the like. What makes a closed form "odd"? What makes this one odd? The juxtaposition of the arccotangent and ϕ? This could lead to...a weird discussion, but a fun one nonetheless. Commented Dec 18, 2013 at 23:41
  • 16
    I know that ϕhas some nice properties, but at the end of the day it's just another algebraic number:
    ϕ=1+52,
    and all we've done is taken the arc cotangent of the square root of it and multiplied by 4π. - just wondering why this result would be seen as "odd". It most likely appears because it happens to be a root of some polynomial which has (1+5)/2as one of its roots.
    Commented Feb 12, 2014 at 18:11
  • 11
    @RonGordon When I see things like this, I don't consider it odd so much as a pleasant surprise. Phi is known as a number of beauty for many reasons, and to see it emerge from, to quote our friend dfeuer, "that horrid-looking thing", is quite a pleasant surprise indeed! No tin-foil hats, no cube-root-of-31-conspiracies, just running into an old friend at the grocery =)
    – corsiKa
    Commented Mar 13, 2014 at 19:28
  • 122
    @corsiKa: ϕalone, sure. But now imagine your meeting your old friend, but dressed in drag with a Kaiser-era military helmet on, spike and all. That's sort of the feeling you get when you see, not regular old ϕ, but ARCCOT SQRT ϕ. Commented Mar 14, 2014 at 17:08
389
+500

I=4πarccotϕ

CC BY-SA 3.0
3
203

NEW ANSWER. I found yet another way of solving this problem. My new solution does not use contour integration, and is based on the following observation: for |z|1,

111x1+x1xlog(1zx)dz=πsin1zπlog(12+121z2).

As I want to keep both the old answer and the new answer, I posted my new solution to other page. You can check it here.


OLD ANSWER. Okay here is another solution. It is also related to my generalization.

We claim the following proposition:

Proposition. If 0<r<1and r<s, then

(1)I(r,s):=111x1+x1xlog(1+2rsx+(r2+s21)x212rsx+(r2+s21)x2)dx=4πarcsinr.

Assuming this proposition, all that we have to do is to solve the non-linear system of equations

2rs=2andr2+s21=2.

The unique solution satisfying the condition of the proposition is r=ϕ1and s=ϕ. So by (1)we have

111x1+x1xlog(1+2x+2x212x+2x2)dx=I(ϕ1,ϕ)=4πarcsin(ϕ1)=4πarccotϕ.

Thus it remains to prove the proposition.


Proof of Proposition. We divide the proof into several steps.

Step 1. (Case reduction by analytic continuation) We first remark that given rand s, we always have

(2)min1x1{1±2rsx+(r2+s21)x2}>0.

Indeed, it is not hard to check if we utilize the following equality

1±2rsx+(r2+s21)x2=(1±rsx)2(1r2)(1s2)x2.

Then (2)shows that the integrand of I(r,s)remains holomoprhic under small perturbation of sin C. So it allows us to extend sI(r,s)as a holomorphic function on some open set containing the line segment (r,)C. Then by the principle of analytic continuation, it is sufficient to prove that (1)holds for r<s<1.

Step 2. (Integral representation of I) Put r=sinαand s=sinβ, where 0<α<β<π2. Then

I(r,s)=111+xx1x2log(1+2rsx+(r2+s21)x212rsx+(r2+s21)x2)dx=012x1x2log(1+2rsx+(r2+s21)x212rsx+(r2+s21)x2)dx( parity)=12x21log(x2+2rsx+(r2+s21)x22rsx+(r2+s21))dx(xx1)=012tlog((t+t1)2+4rs(t+t1)+4(r2+s21)(t+t1)24rs(t+t1)+4(r2+s21))dt,

where in the last line we utilized the substitution x=12(t+t1). If we introduce the quartic polynomial

p(t)=t4+4rst3+(4r2+4s22)t2+4rst+1,

then by the property p(1/t)=t4p(t), we can simplify

I(r,s)=201logp(t)logp(t)tdt=0logp(t)logp(t)tdt=0(p(t)p(t)+p(t)p(t))logtdt=12(p(z)p(z)+p(z)p(z))logzdz,

where we choose the branch cut of login such a way that it avoids the upper-half plane

H={zC:z>0}.

Step 3. (Residue calculation) Since

f(z):=(p(z)p(z)+p(z)p(z))logz=O(logzz2)as z,

by replacing the contour of integration by a semicircle of sufficiently large radius, it follows that

I(r,s)=12{2πiz0HResz=z0f(z)}=πz0HResz=z0f(z).

(It turns out that f(z)has only logarithmic singularity at the origin. So it does not account for the value of I(r,s).) But by a simple calculation, together with the condition 0<α<β<π2, we easily notice that the zeros of p(z)are exactly

e±i(α+β)ande±i(αβ).

Now let Z+be the set of zeros of p(z)in Hand Zbe the set of zeros of p(z)in H. Then

Z+={ei(β+α),ei(βα)}andZ={ei(β+α),ei(βα)}.

This in particular shows that

p(z)p(z)logz=z0Z+logzzz0+holomorphic function on H

and

p(z)p(z)logz=z0Zlogzzz0+holomorphic function on H.

So it follows that

I(r,s)=π{z0Z+logz0z0Zlogz0}=π{logei(β+α)+logei(πβ+α)logei(πβα)logei(βα)}=π{i(β+α)+i(πβ+α)i(πβα)i(βα)}=4πα=4πarcsinr.

This completes the proof.

CC BY-SA 3.0
3
  • 12
    Very nice, and it helps me simplify a piece of my proof as well (I think our solutions have more in common than not). One question, though: what about the branch point of the log in the residue calculation? I know it doesn't seem to matter as you do end up with the correct solution, but you may want to say something about avoiding the branch point at the origin and defining a branch of the log (which I think you do anyway with your restrictions on αand β) in the complex plane. Commented Nov 17, 2013 at 8:45
  • 5
    @RonGordon, As written in the solution, the branch cut of the log is chosen so that it avoids the upper half plane. So it would be safe if we choose it as the negative y-axis, but I think the standard branch cut (,0)would also works. Commented Nov 17, 2013 at 15:44
  • 3
    Oh my, there it is. My bad, so sorry. In any case, reading your solution helped me simplify a small part of mine, so thanks. Commented Nov 17, 2013 at 15:53
48

Our aim is to give an elementary proof of Proposition formula (1) in the answer of @sos440. We first note that

min1x1{1±2rsx+(r2+s21)x2}>0.
Indeed, if x=±1then
1±2rsx+(r2+s21)x2(rs)2>0,
if x=0then
1±2rsx+(r2+s21)x2=1>0,
if 1<x<1, x0then the equations
s(1±2rsx+(r2+s21)x2)=0,r(1±2rsx+(r2+s21)x2)=0,
give ±r=sx, ±s=rx, which is impossible.

In the second step we show that I(r,s)is independent of s.

sI(r,s)=111+x1x4r(1+(r2s21)x2)(12rsx+(r2+s21)x2)(1+2rsx+(r2+s21)x2dx.
Substituting x:=xand adding them we obtain
2sI(r,s)=1121x24r(1+(r2s21)x2)(12rsx+(r2+s21)x2)(1+2rsx+(r2+s21)x2dx,
that is,
sI(r,s)=1111x24r(s2+r21)x2+4r1+(r2+s21)2x4+(2s24r2s2+2r22)x2dx.
Substituting x:=sin(t)we have
sI(r,s)=π/2π/24r(s2+r21)sin(t)2+4r1+(r2+s21)2sin(t)4+(2s24r2s2+2r22)sin(t)2dt
=π/2π/28r((s2+r21)cos(2t)+s2r21)(r2+s21)2cos(2t)22(r2s21)(r2+1s2)cos(2t)+r4+(26s2)r2+(s2+1)2dt
=ππ4r((s2+r21)cos(y)+s2r21)(r2+s21)2cos(y)22(r2s21)(r2+1s2)cos(y)+r4+(26s2)r2+(s2+1)2dy.
Introducing the new variable T:=tany2we obtain
sI(r,s)=4r(s2r2)T24r(rs)2(r+s)2T4+((24s2)r2+2s2)T2+1dT=4r(s2r2)(rs)2(r+s)2T2+aT4+bT2+b2/4+ddT=4r(s2+r2)(rs)2(r+s)2(2a(b2+4d)+(b2+4d)3/2)π(b2+4d)3/2b2+4d+b,
where
a=1s2r2,
b=(24s2)r2+2s2(rs)2(r+s)2,
b2+4d=4(rs)2(r+s)2.
It gives 2ab2+8da+(b2+4d)3/2=0.

Since sI(r,s)=0we have

I(r,s)=I(r,1)=111x1+x1xlog((1+rx)2(1rx)2)dx.
From this
rI(r,1)=111+x1x41r2x2dx.
Similarly as above we get
rI(r,1)=1141x2(1r2x2)dx=4π1r2=4π(arcsinr).
It implies
I(r,1)=4πarcsinr+C.
Taking the limit limr0+we obtain C=0, that is, I(r,s)=4πarcsinr.

CC BY-SA 3.0
2
44

For the purposes of alternative methods, it may be of interest to note that the integrand

f(x)=1x1+x1xlog(2x2+2x+12x22x+1)
may be rewritten in terms of hyperbolic trigonometric functions. Using
tanh1(z)=12log(1+z1z),
and we obtain

f(x)=1xetanh1xlog(1+2x1+2x212x1+2x2)=etanh1x(2tanh1(2x1+2x2)x).

The rational function in the bracket, which we will denote s(x), is symmetric about x=0.

The desired integral is

I=11f(x)dx=11etanh1xs(x)dx,

which, by adding the indicated useful definite integral to both side, gives

I+11etanh1xs(x)dx=211s(x)dx1x2.

Now using the change of variable x=ywe have

11etanh1xs(x)dx=11etanhys(y)dy=11etanhys(y)dy=I,
by the symmetry of s(x). Hence, we finally obtain

I=11s(x)dx1x2=2111x1x2tanh1(2x1+2x2)dx.

This integral is symmetric about x=0, so we have

I=4011x1x2tanh1(2x1+2x2)dx,
which can be rewritten
I=401(ddxsech1x)tanh1(2x1+2x2)dx.

Using integration by parts this results in

I=801sech1(x)(12x2)1+4x4dx.


We could also make the change of variable y=sech1xto obtain

I=80y(cosh2(y)2)sinhycosh4(y)+4dy=80ysinh3ycosh4y+4dy80ysinhycosh4y+4dy.

CC BY-SA 4.0
1
27

This answer provides a way to find I=01ln(x42x2+5)ln(5x42x2+1)1x2 dx(which @RonGordon obtained above) with differentiating under the integral sign. A u-substitution of u=1+x21x2yields this.

I=121ln(u2+2u+2u22u+2)u21 du.
Now integrate by parts with a=ln(u2+2u+2u22u+2)and db=duu21.
I=ln(u2+2u+2u22u+2)ln(u+u21)]1+21u22u4+4ln(u+u21) du
The first term is equal to 0, so we are left with this.
I=21u22u4+4ln(u+u21) du
We now begin the step of differentiating under the integral. Consider the following integral:
f(a)=a1x2a2x4+a4ln(x+x21) dx
Note that trivially, f(0)=0.A quick u=xayields this.
f(a)=1au21u4+1ln(au+(au)21) du
Differentiating with respect to aand using the Chain Rule, we get this.
f(a)=1×1a2×(1a)21(1a)4+1ln(a(1a)+(a(1a))21)+1ax21x4+1×x(ax)21 dx
Luckily, the first term cancels, so we are left with this.
f(a)=1ax21x4+1×x(ax)21 dx
A u-substitution of u=(ax)21yields this.
f(a)=0u2+1a2(u2+1)2+a4 du
Consider the integral with ua4+1u
f(a)=1a4+10(1a2)u2+(a4+1)u4+2u2+(a2+1) du
If we add these two versions of the integral and divided the numerator and denominator of the integrand by u2, we get the following.
f(a)=(1a2)+a4+12a4+1×01+a4+1u2(ua4+1u)2+2(1+a4+1) du
We can finally perform a very nice substitution of w=ua4+1uto solve this integral.
f(a)=(1a2)+a4+12a4+1×dww2+2(1+a4+1) dw
Thus, we can finally say that f(a)=(1a2)+a4+12a4+1×π2(1+a4+1).After a bit of considerable algebra, we can simply that to obtain this.
f(a)=π2a4+1a2a4+1
Integrating, we can now say this about the value of f(a).
f(a)=π20ax4+1x2x4+1 dx
Only one u-substitution of u=x4+1x2is required here to obtain this.
f(a)=π22a4+1a21du1u2
This, of course, is equal to πarccos(a4+1a2)22.

We will now manipulate this result to a function with arctanin it.

f(a)=πarccos(a4+1a2)22=π2arctan(a4+11a2)

Our desired value for our original integral is 2f(2).

01ln(x42x2+5)ln(5x42x2+1)1x2 dx=πarctan(512)=πarccotϕ

So the final answer to the original problem is 4πarccotϕ.

CC BY-SA 3.0
4
22

Noteworthy, RIES (http://mrob.com/pub/ries/index.html) finds closed form from numerical value in the form of an equation:

cos(xπ)+1=2ϕ6.

Simplifying above, we get another form of the result:

I=πarccos(1785).

CC BY-SA 3.0
3
20

Figured I would contribute and add a self-contained real analytic method:

I will use the following representations that are fairly straight-forward to prove:

20cos(x)cos(tx)x(exabi+exa+bi)dx=ln((t2a)2+b2)ln((1a)2+b2)
0sin(x)x(exabi+exa+bi)dx=arctan(1abi)+arctan(1a+bi)
0cos(t)cos(tx)x21dx=π2sin(t),fort0

Now we can begin evaluating I:

I=111x1+x1xln(2x2+2x+12x22x+1)dx

Enforce the substitution 1+x1x=u:

I=04u2u41ln(5u42u2+1u42u2+5)du=u1u04u41ln(5u42u2+1u42u2+5)du

From adding the two, we can deduce:

I=02u21ln(5u42u2+1u42u2+5)du

Since

ln(5u42u2+1)ln(4)=ln((u215)2+425)ln(45)
ln(u42u2+5)ln(4)=ln((u21)2+4)ln(4)

We can write

A=02u21(ln((u215)2+425)ln(45))duB=02u21(ln((u21)2+4)ln(4))du
such that I=AB

Now from the starting representations we deduce:

A=04u210cos(t)cos(ut)t(et(12i)/5+et(1+2i)/5)dtdu=2π0sin(t)t(et(12i)/5+et(1+2i)/5)dt=2πarctan(512i)+2πarctan(51+2i)=(4πarctan(512i))

Similarly, one finds

B=(4πarctan(112i))

By using the identity:

arctan(x)+arctan(y)=arctan(x+y1xy)

One deduces

I=4πarccot(ϕ)

CC BY-SA 4.0
    15

    Well, 9 years later, this answer is going to utilize an approach via a contour integral. Denote

    I=111x1+x1xln(2x2+2x+12x22x+1)dxR
    our integral of interest.

    Complex function

    Let

    f(z)=1zz+1z1(ln(2z+1+i2z1+i)+πi2),
    where wand lnware chosen such that they have a branch cut at the negative real axis (the so-called principal branches) in w-plane, that is argw(π,π]. We denote
    h(z)=1zz+1z1,g(z)=ln(2z+1+i2z1+i)+πi2
    so f(z)=h(z)g(z). To find the branch cuts of f(z), we need to find which points on the z-plane are mapped onto the negative real axis in w. That is to solve

    wB.C.:w=z+1z1=t,t>0;lnwB.C.:w=2z+1+i2z1+i=t,t>0.
    After simple manipulations, we arrive at
    h(z)B.C.:z=t1t+1(1,1);g(z)B.C.:z=t12(t+1)i2(1i2,1i2).
    where (z1,z2)means here a line segment from z1to z2for any in general complex z1,z2.

    Poles and residues

    At z, we have f(z)=πi2z+O(1/z2), a simple pole. So

    Resf(z)=limzzf(z)=πi2.
    Note that there is no pole at z=0since
    2z+1+i2z1+i=i2iz+O(z2),
    so
    g(z)=πi2+2z+πi2+O(z2)=2z+O(z2),
    which cancels the simple pole of h(z)at 0±0i.

    Branch jumps

    • (1,1):Let t(1,1). Since the complex square root function changes sign on the branch discontinuity, h(t+i0)=h(ti0)and so f(t+i0)=f(ti0).
    • (1i2,1i2):Let t(1/2,1/2). For z=t1/2i±i0,
      2z+1+i2z1+i|ti/2±i0=r+(2z+1+i2z1+i)|ti/2(±i0)=r4(12t)2(±i0)=ri0.
      where ris some negative number (branch cut mapped on negative numbers r). Hence,
      g(t+i0)g(ti0)=ln(ri0)ln(r+i0)=2πi.

    Residue theorem

    We integrate f(z)along contour Cdefined in Figure below. The contour is also split as C=C1++C6.

    Contour of integration

    According to the Residue theorem,

    Cf(z)dx=2πiResf(z)=π2.

    On the other hand,

    (1)Cf(z)dz=k=16Ckf(z)dz.

    Parametrisation

    For C1,C6, we have the following parametrisation:

    • C1:z=t+i0,t(1,1),dz=dt; Since h(z)has a branch cut here, f(t+i0)=h(t+i0)g(t). To obtain h(t+i0), we first compute
      z+1z1|t+i0=z+1z1|t+(z+1z1)|ti0=1+t1t2(1t)2i0=1+t1ti0,
      so h(t+i0)=it1+t1t. For g(z), we can write
      g(t)=ln|2t+1+i2t1+i|+arg(2t+1+i2t1+i)+πi2=12ln(2t2+2t+12t22t+1)+iξ(t)
      for some real function ξ(t)which vanishes at t=0. Although we don't need this function, we express it anyway for completeness: Since π<arg(2t+1+i)arg(2t1+i)<π, we have arg(2t+1+i2t1+i)=arg(2t+1+i)arg(2t1+i)and so
      ξ(t)=arctan(2t1)arctan(2t+1)+π2.
      Overall,
      f(t+i0)=it1+t1t(12ln(2t2+2t+12t22t+1)+iξ(t)).
      Hence
      C1f(z)dz=C1f(z)dz=11f(t+i0)dt=i2II0,
      where we denote
      I0=11ξ(t)t1+t1tdtR.
    • C2: Using knowledge of branchjumps,
      C2f(z)dz=C2f(z)dz=C1(f(z))dz=i2II0.
    • C3=C6, so
      C3f(z)dz+C6f(z)dz=C3f(z)dzC6f(z)dz=0
    • C4: Using knowledge of branchjumps,
      C4f(z)dz=C4h(z)g(z)dz=C5h(z)(g(z)2πi)dz=C5f(z)dz+2πiC5h(z)dz.

    Comparison & Antiderivative

    Comparing the terms, that is by using (1), we get

    (2)π2=iI2I0+2πiC5h(z)dz.

    We now solve the remaining integral C5h(z)dz. For brevity, we denote a=1i2,b=1i2, so C5=(a,b). Notice h(z)is now holomorphic in the vicinity of C5. If we find its antiderivative H(z)also holomorphic there, then

    (3)C5h(z)dz=H(b)H(a).

    This task is rather simple. First, we start by integrating h(z)as if it was a real function. For x>1, we have by substitution (x1)/(x+1)=u2,

    h(x)dx=1xx+1x1dx=4u2u41du=2argcoth(x+1x1)+2arccot(x+1x1).

    Hence, a suitable antiderivative in the complex plane is the following:

    H(z)=ln(z+1z1+1z+1z11)iln(z+1z1+iz+1z1i).

    We know the branch of H(z)due to ν=z+1z1is z(1,1). Assuming the principal branches, we now check where are the other possible branches of H(z)due to ln's.

    lnwB.C.:w=ν+1ν1=t,t>0;lnwB.C.:w=ν+iνi=t,t>0.
    w(ν)B.C.:ν=t1t+1=∈(1,1);w(ν)B.C.:ν=it1t+1(i,i).
    H(z)B.C.:z=ν2+1ν21(,0).

    Hence, the B.C. of H(z)lies at (,1), safely away from C5. Substituting (3)into (2), we get the value of I. And as a bonus, also I0. For I,

    (4)I=2πRe(H(a)H(b)).

    Substituing aand binto z+1z1, we get

    a+1a1=15+2i5=1q2e2αi,b+1b1=1+2i=q2e2αi
    with q=51/4>1and α=π212arctan2. Note that π/4<α<π/2, so

    a+1a1=1qeαi,b+1b1=qeαi.

    In the case of b, substituing into ν=z+1z1and after simple manipulations,

    ν+1ν1|ν=qeαi=q22iqsinα1q22qcosα+1,ν+iνi|ν=qeαiq2+2iqcosα1q22qsinα+1.

    Hence, since the denominators are real,

    H(b)=ln(q22iqsinα1)ln(q22qcosα+1)iln(q2+2iqcosα1)+iln(q22qsinα+1),

    expanding the logarithms into real and imaginary part,

    H(b)=12ln(q42q2cos(2α)+1)+iarg(q22iqsinα1)ln(q22qcosα+1)12iln(1+2q2cos(2α)+q4)+arg(q2+2iqcosα1)+iln(q22qsinα+1)

    Since we are interested only in real part (just I),

    ReH(b)=12ln(q42q2cos(2α)+1)ln(q22qcosα+1)+arg(q2+2iqcosα1)

    Similarly for a(change qto 1/q).

    ReH(a)=12ln(1q42q2cos(2α)+1)ln(1q22qcosα+1)+arg(1q2+2iqcosα1)=12ln(q42q2cos(2α)+1)ln(q22qcosα+1)+πarg(q2+2iqcosα1).

    Therefore, we have ourselves the grand finale,

    I=4π(π2arg(q2+2iqcosα1))=4πarccot2qcosαq21.

    To show this result is the same as the ones found already by others, note that

    (2qcosαq21)2=4q2sin2(π2α)(q21)2=45sin2(12arctan2)(51)2=25115(51)2=1+52.

    CC BY-SA 4.0
    1
    14

    This is not really an answer, but grossly too long for an comment. I didn't know how to simplify it beyond the final solution.

    I=111x1+x1xln(2x2+2x+12x22x+1)dx

    Begin with the substitution of x=cos2a

    I=111cos2a1cos2a1+cos2aln(2cos22a2cos2a+12cos22a2cos2a+1)dx

    By the tangent and cos double angle properties

    I=11sec2a|tana|ln(2cos22a+cos4a+22cos2a+cos4a+2)da

    Were just getting started. Now replace a=12gd(b)where gdis the Gudermannian function.

    I=11sec(gd(b))|tan(gd(b2))|ln(2cos2(gd(b))+cos(2gd(b))+22cos2(gd(b))+cos(2gd(b))+2)da

    Hehe. Now we get to simplify a bit. This is under the definition of Gudermannian properties.

    I=11cosh b|sinhb2|ln(2sech2b+(sech2b+tanh2b)+22sech2b+(sech2b+tanh2b)+2)

    Now, use properties of tanhand sechto simplify even further

    I=11cosh b|sinhb2|ln((1sech2b)+2(1+sech2b)+2)

    Our goal is to create an arctanhfunction, but that will obviously take some serious effort. Factor out a 3to generate that 1needed even if it makes an ugly factoring.

    I=11cosh b|sinhb2|ln(3(1sech2b3)3(1+sech2b3))

    And now cut out all of the 3's. After this cut, use a property of ln's to reciprocate the argument of ln. And multiply 2 and 1/2

    I=112cosh b|sinhb2|12ln((1+sech2b3)(1sech2b3))

    And what do you know! You're there! Use a property of lnand arctanhto generate a much CLEANER form (also by throwing the 2 in front).

    I=211cosh b|sinhb2|arctanh(sech2b3)

    This function is even, and we can know that because all parts of what is above, coshb,|sinhb|,etc. all even. So we can do the following.

    I=401cosh b|sinhb2|arctanh(sech2b3)

    This is just an idea, and like I said not a real solution. I have no idea where to continue beyond this, but I thought it may help to come up with a new idea to solve.

    CC BY-SA 3.0
    3
    • 2
      After further inspection, I messed up my work here. I will leave this post here howver becUse the purpose of the post still holds (ideas to solve)
      – user285523
      Commented Nov 22, 2015 at 17:42
    • Don't you need to change the limits after you make the first change of variable x=cos2a?
      – r9m
      Commented Dec 2, 2015 at 16:59
    • @user23055 not really. There are lots of mistakes and only consists of only substitution
      – user311151
      Commented Apr 2, 2016 at 4:48
    13

    I have a relatively straightforward elementary approach.

    First you use the substitution xxand add the integrals to obtain

    I=111x11x2ln(2x2+2x+12x22x+1) dx.

    A further substitution, x2x1+x2yields

    I=111xln(4x2+(x+1)44x2+(x1)4) dx=2111xln(4x2+(x+1)4) dx.

    Define the function J by

    J(λ)=11ln((x+1)2+2iλx)xdx.
    Then
    I=2(J(1)+J(1)).

    Differentiating under the integral sign gives

    J(λ)=2i111((x+1)2+2iλx)dx=iπλ22iλ=iπ1(1+iλ)2.

    Hence

    J(λ)=πarcsin(1+iλ)+C.
    But
    J(0)=211ln(x+1)xdx=π2/2
    which implies C=0, hence
    I=2π(arcsin(1+I)+arcsin(1I))=4πarcsin(1ϕ)=4πarccotϕ
    using standard identities.

    CC BY-SA 4.0
      13

      Version of 17.09.23.

      Let y=1+x1x,then x=1y21+y2,

      I=111x1+x1xln2x2+2x+12x22x+1dx=40lny42y2+55y42y2+1dy1y4=I0I1,
      where
      (1)I0=20lny42y2+55y42y2+1dy1y2,I1=20lny42y2+55y42y2+1dy1+y2.
      By the substitution y=1zeasily to get
      1lny42y2+55y42y2+1dy1+y2=01lnz42z2+55z42z2+1dz1+z2.
      I.e.
      (2)I1=0.

      On the other hand,

      I0=I00I01,
      where
      (3)I00=20lny42y2+5(y2+1)2dy1y2,I01=20ln5y42y2+1(y2+1)2dy1y2,
      By the substitution y=1zeasily to get
      0ln5y42y2+1(y2+1)2dy1y2=0lnz42z2+5(z2+1)2dz1z2,
      (4)I01=I00,I=2I00,

      I=2I00=40lny42y2+5(y2+1)2dy1y2
      =40ln(y212i)+ln(y21+2i)2ln(y2+1)1y2dy
      (5)I=4(J(12i)+J(1+2i)2J(1)),
      where
      (6)J(p)=0ln(y2+p)1y2dy,
      dJ(p)dp=0dy(y2+p)(1y2)=F1+F2(p)1+p,
      F1=011y2dy,F2(p)=01y2+pdy.
      By the substitution y=1teasily to get
      0111y2dy=111t2dt,F1=0.
      At the same time,
      F2(p)=1parctanyp|0=π2p.
      Therefore,
      dJ(p)dp=π2(1+p)p,
      J(p)=π1+pdp2p=πarctanp+C.
      From (5)should
      I=4π(arctan12i+arctan1+2i2arctan1)=4π(arctan12i+1+2i15π2)=4π(arctan2(51)15π2)=4π(arctan(5+12)π2)=4π(πarctan5+12π2),I=4πarccotφ8.37221162660127566162574712109841263808172805388220741371709,
      where φ=5+12is the golden ratio.

      CC BY-SA 4.0
      3
      • 2
        Nice solution. Although are you sure the wolfram alpha result is correct? For positive real a, it seems the integral should return a real value (assuming the Cauchy principal value), which means the imaginary term probably shouldn't be there. Wolfram also says the answer is for Re(a)>0 which is not that case for a = -1 +/- 2i. Maybe you could try Feynman integration if you want to do it explicitly. Or just look at the explicit antiderivative from Wolfram and treat it carefully.
        – Andrew
        Commented Jun 23, 2023 at 7:48
      • @Andrew Thank you for the deep comment! Honestly, I was not sure in the correctness of this integral too. Undoubtedly, if 𝑎 is real, then its negative value leads to the singularity of ln(z^2+𝑎) in the domain of the integration. At the same time, this problem does not exist in the task conditions. Commented Jun 25, 2023 at 9:16
      • @Andrew New version (by your notes) is ready. Commented Sep 17, 2023 at 15:32
      12

      Eight years later.

      Starting from @Ron Gordon's substitution

      80(u21)(u46u2+1)u8+4u6+70u4+4u2+1log(u)du
      since the roots of the polynomials in u2are simple, we can use partial fraction decomposition (I shall not type the formulae) and we face four integrals
      I=αβx2+γlog(x)dx
      where all coefficients are complex numbers. Then
      I=iα(Li2(ixβγ)Li2(ixβγ)+log(x)(log(1iβxγ)log(1+iβxγ)))2βγ
      which make
      J=0αβx2+γlog(x)dx=iα(log2(iβγ)log2(iβγ))4βγ=παlog(βγ)4βγ
      This gives as a result
      80(u21)(u46u2+1)u8+4u6+70u4+4u2+1log(u)du=π(πcot1(1422+175))
      I have not been able to simplify further.

      Edit

      If you look at this question of mine, @Jyrki Lahtonen made the simplification I was not able to do.

      CC BY-SA 4.0
        7

        Since this question received a lot of answers recently, I thought I might partake as well. My solution to this historical question (on this site, at least) requires a lemma, namely:

        0dx2+eax4+bx2+cdx=π22ac+b[da+ec]
        This works for a,b,c>0,d,eR. (the variable range can be larger, but this is enough for this question.) I shall provide a proof, too. The first time I had derived this was through complex analysis, but here I will do it more elementarily.
        To start with, assume fis a even real function, then,
        f(xp2q2x)dx=0f(xp2q2x)dxx=pqet+0f(xp2q2x)dxx=pqet=f(qp(etet))pqetdt+f(qp(etet))pqetdt=f(2qpsinh(t))2pqcosh(t)dt=2qpsinh(t)=xp2f(x)dx
        So we have
        0dx2+eax4+bx2+cdx=12dx2+eax4+bx2+cdx=d2x2ax4+bx2+cdx+e21ax4+bx2+cdx=d2x2cx4+bx2+adx+e2x2ax4+bx2+cdx=d21cx2+b+ax2dx+e21ax2+b+cx2dx=d21(cxax)2+2ac+bdx+e21(axcx)2+2ac+bdx=d2c1x2+2ac+bdx+e2a1x2+2ac+bdx=π22ac+b[da+ec]


        Now, let's prove the integral at hand! Actually, let's prove a broader case. Given r>|s|,

        111x1+x1xln((2r1)x2+2sx+1(2r1)x22sx+1)dx=4πsgn(s)arcsin(rr2s2)
        notice that I(r,0)=0and I(r,s)=I(r,s), therefore we can only consider the case where s>0
        I(r,s)=111x1+x1xln[(2r1)x2+2sx+1(2r1)x22sx+1]dx=2011x1x2ln[(2r1)x2+2sx+1(2r1)x22sx+1]dx=(1)20π21cos(θ)ln[(2r1)cos2(θ)+2scos(θ)+1(2r1)cos2(θ)2scos(θ)+1]dθ=(2)40111t2ln[(2r1)(1t2)2+2s(1t2)(1+t2)+(1+t2)2(2r1)(1t2)22s(1t2)(1+t2)+(1+t2)2]dt
        Then differentiate
        sI(r,s)=401[1+t2(rs)t4+(22r)t2+(r+s)+1+t2(r+s)t4+(22r)t2+(rs)]dt=(3)41[1+t2(r+s)t4+(22r)t2+(rs)+1+t2(rs)t4+(22r)t2+(r+s)]dt=20[1+t2(r+s)t4+(22r)t2+(rs)+1+t2(rs)t4+(22r)t2+(r+s)]dt=(4)2π(rs+r+s)r2s22r2s2+22r
        Integrate again
        I(r,λ)=0λsI(r,s)ds=2π0λrs+r+sr2s2r2s2+1rds=(5)4π2r0X1(t2+1)(12r)t2+1dt=(6)4π0tan(X)2rcos(θ)12rsin2(θ)dθ=4πarcsin(2rsin(tan(X)))=4πarcsin(rr2λ2)
        (1) x=cos(θ)
        (2) tan(θ2)=t
        (3) t1t
        (4) Use the lemma
        (5) s=2rt1+t2
        (6) t=tan(θ)
        (7) X=rr2λ2λ


        So, for this specific question, we have r=1.5,s=1. After plugging these values in, we got

        111x1+x1xln(2x2+2x+12x22x+1)dx=4πarcsin(352)=4πarccot(1+52)

        CC BY-SA 4.0
        4

        You must log in to answer this question.

        Not the answer you're looking for? Browse other questions tagged.