Hey /sci/, this is a problem I found on another forum which stumped

This is the function f(x) = x^x , x = (0,1), in case you need it...

https://www.wolframalpha.com/input/?i=x%5Ex+from+0+to+1

well if a=b that solves the "equals to" portion of the problem, and if a>b then a^a will always be greater than a^b and since its all exponentials b^a will never be able to pick up the slack from the a^b term to make up the difference, even if a is large

>>8779013
> if a>b then a^a will always be greater than a^b if a>b then a^a will always be greater than a^b

That is incorrect, if a>b then a^a will always be SMALLER than a^b, because since b is smaller, then a^b will be closer to 1 than a^a.

Also what do you mean by pick up the slack, could you elaborate on that please?

a>b => a^a>b^a and b^b>a^b => a^a+b^b > a^b+b^a

>>8779021
a>b does NOT imply b^b>a^b

If a>b, then from order from biggest to smallest is:

1. (a^b)
2. (a^a) & (b^b) (depends on the values of a&b)
3. (b^a)

i.e, the smaller the exponent, the bigger the term

>>8778990
I don't know about a formal proof, but here is an intuition about why this is correct, maybe it'll help you with the proof itself:
Let's assume that a is larger than b. If it helps you, imagine that a is much larger than b (e.g a is 0.9, b is 0.1).
a is closer to 1 than b, so it is less affected by exponents. No matter what power you raise it to, b will be more affected by that same power (because it is closer to 0). This also means that if you want to make you sum as big as possible, you shouldn't put too much "work" in making a larger, because it won't make it much bigger.
Instead, you should "put most of your money" on b. In terms of exponentiation, this means that you should give it the more effective power out of the two - i.e the smaller one (because smaller powers affect numbers more powerfully).
So you should give a the power a, because it won't be affected no matter what power you give it. And you should also give b the power b, because the smaller the power, the larger it will get (and like I said, b will get larger faster than a).
Hope this helps.

>>8779054
Ok I understand now (intuitively at least).
So in summary:
if (a>>b)
then b^b-b^a>=a^b-a^a is intuitive, because b is smaller and therefore more affected, hence the difference between the b's is bigger than the difference between the a's. Thanks!

File: Screen Shot 2017-03-26 at 1.42.42 PM.png (1MB, 2092x1416px) Image search: [Google]

1MB, 2092x1416px

Ran a quick simulation myself because I am autistic

df<- data.frame(index = c(1:10000), a=runif(10000, 0, 1), b = runif(10000, 0 , 1))
head(df)
df<-df%>%
mutate(lhs = a^a+b^b, rhs = a^b+b^a, check = lhs>rhs)
length(which(df$check == FALSE))

ggplot(df, aes(x = index))+
geom_point(aes(y = lhs), color = "firebrick", size = 1/50)+
geom_point(aes(y = rhs), color = "deepskyblue3", size = 1/50)

File: Screenshot from 2017-03-26 13-55-03.png (140KB, 620x490px) Image search: [Google]

140KB, 620x490px

Just attach this image.

>>8778990

ok this problem seems fun!

i've tried working out with symmetry only using "half-angle", turns out (a+b)/2 and (a-b)/2 seem like good ideas but i'm stuck at the question
[math](a/b)^{\frac{a+b}{2}} \leq \frac{b^{\frac{a-b}{2}} - b^{\frac{b-a}{2}}}{a^{\frac{a-b}{2}} - a^{\frac{b-a}{2}}}[/math]?
even taking a look at the behavior of x-1/x (quotient of two of these on the right)

i need to go to bed, but you might want to try expressing one of the functions f(a,b) = a^a + b^b - a^b - b^a, or g(a,b) = (a^a + b^b)/(a^b + b^a) in terms of the symmetric variables s = a+b and d = a-b (maybe halved) and study the partial derivatives along d==0 axis...

i'll try again tomorrow, and hope you find before me so i can have that good feeling i get when i check my work and realise someone else's mind seem to agree with mine :)

have fun

File: SUCCtion.jpg (105KB, 1280x738px) Image search: [Google]

105KB, 1280x738px

For a fixed [math]b \in (0, 1)[/math], define [math]f \colon (0, 1) \to \mathbb{R}[/math] by [math]x \mapsto x^x + b^b - x^b - b^x[/math]. This is continuous, and [math]f'(x)=(\text{ln } x + 1)x^x -bx^{b-1} -b^x \text{ln }b[/math] has a root at [math]x=b[/math]. If you could check the behaviour of [math]f'[/math] on both sides of [math]b[/math], you would, very likely, find out it's where you get an extremum of the function. Then, you'd see that [math]f(b)=0[/math], and use this to get the desired result. Details are left for the reader.

I don't remember which inequality it was (could have been precisely this), but this is how I got the proof done.

>>8781330
That's what I would have done, but how do you show that b is a global minimum? You'd need to show that f'(x) has only one root at x = b (which it presumably does), but that would involve a lot more non-trivial work

File: 1484085372421.jpg (109KB, 1280x720px) Image search: [Google]

109KB, 1280x720px

>>8781464
How's the second derivative? If it has the same sign, or, equivalently, is non-zero everywhere on this interval, then you get the uniqueness out of that.

I wonder what the inequality was I proved, it wasn't this. >>8781330 and some fixing here, continuous -> differentiable.

>>8778990
take logs of both sides, subtract Rhs from Lhs, differentiate and investigate the first two derivatives

File: brainlets.jpg (93KB, 639x1136px) Image search: [Google]

93KB, 639x1136px

>>8778990

Saw this post yesterday, remembered I did this problem once in high school, but had completely forgotten since.

I'm doing a PhD in applied math now, so I asked my colleagues at lunch, including professors, and finally after 3 hours, a female PhD student came up with an elementary proof. Guess we're all brainlets lol.

Joking aside, this highlights quite well the difference between being good at math contest style problems and working in math.

>>8782907
I can't barely read the handwriting.

>>8784470

sorry, blame the thot that wrote it, also it's in French, but I'm too lazy to rewrite it, anyhow you can pretty much figure out everything from the math parts, she just used convexity and monotonicity, pretty elegant actually

>>8782907
"Revient à"
Can't you write some <=> like everyone ?

>>8785188

again, it was a chick that wrote this, I was too lazy to rewrite/translate it, since from a mathematical point of view, everything is there, so problem solved

File: HETLXKv.jpg (902KB, 2390x4250px) Image search: [Google]

902KB, 2390x4250px

>>8778990
r8 h8 masturb8

I noticed after the fact that this proof also shows that the inequality is strict only when a = b

>>8786067
I meant attained, not strict

>>8784688
>convexity and monotonicity
...of [math]e^x[/math] and [math] \ln x[/math], respectively

the first ratio can be written
[math] \displaystyle \frac{a^a+b^b}{a^b+b^a}= \lambda e^{x_1}+(1- \lambda)e^{x_2}[/math]
where
[math] \displaystyle \lambda= \frac{b^a}{a^b+b^a}[/math]
and
[math] \displaystyle x_1= \left( \frac{a}{b} \right)^a=e^{a( \ln a- \ln b)}[/math]
and
[math] \displaystyle x_2= \left( \frac{b}{a} \right)^b=e^{b( \ln b- \ln a)}[/math]
then the result from convexity of [math]e^x[/math] is a bit simpler.

>>8778990

I tried to put in the most simple way I could:

Let [math] c = max{a,b} , d = min{a,b} [/math]

[math] \frac{c^c+d^d}{c^d + d^b} geq \frac{c^c+d^d}{d^c} = \left(\frac{c}{d}\right)^c + \frac{d^d}{d^c} [/math]

the first parcel is always greater than one.

If a==b, then, the quotient is equal to one.

Cheers

>>8786888
sorry, [math] c = max\{a,b\}, d = min\{a,b\} [/math]

and [math] \frac{c^c+d^d}{c^d+d^c} \geq \left(\frac{c}{d}\right)^c + \frac{d^d}{d^c} [/math]

>>8786888

But in your geq statement, c^d + d^c > d^c. Dividing by a bigger amount always yields a smaller number, not a bigger number.

>>8782907
Quelle école ?

not rigorous but i think this would solve the problem quite elegant
[math]
a^a-a^b+b^b-b^a\ge 0 \\
\Leftrightarrow \int_{b}^{a} {d\over dx} a^x + \int_{a}^{b} {d\over dx} b^x \,dx \ge 0 \\
\Leftrightarrow \int_{b}^{a} {d\over dx} a^x-b^x \,dx\ge 0 \\
\Leftarrow a^x \log a-b^x \log b\ge 0 \\
[/math]
the last part is easy if you assume [math]a>b[/math] and realize [math]x>0[/math]

>>8787354
ok turns out this doesnt work for values of b smaler than 0.03
for example 0.8^7 log 0.8-0.01^7 log 0.01 is negative