Minkowski Inequality

Algebra Level 5

a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 \large \sqrt[3]{a^3+\dfrac{1}{b^3}}+\sqrt[3]{b^3+\dfrac{1}{c^3}}+\sqrt[3]{c^3+\dfrac{1}{a^3}}

Given that a , b , c a,b,c are positive reals satisfying a + b + c 3 2 a+b+c\le \frac{3}{2} . Determine the minimum value of the expression above to 3 decimal places.

A solution using Minkowski Inequality will be very much appreciated!


The answer is 6.031.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

6 solutions

I will be trying to post a Minkowski inequality solution, till then lets look at this.

f ( a , b , c ) = a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 subject to a + b + c 3 2 f(a,b,c) = \sqrt[3]{a^3+\dfrac{1}{b^3}}+\sqrt[3]{b^3+\dfrac{1}{c^3}}+\sqrt[3]{c^3+\dfrac{1}{a^3}} \text{ subject to } a+b+c\le\frac{3}{2}

Using partial differentiation i.e solving f ( a , b , c ) a , f ( a , b , c ) b , f ( a , b , c ) c = 0 \frac{\partial{f(a,b,c)}}{\partial{a}} , \frac{\partial{f(a,b,c)}}{\partial{b}},\frac{\partial{f(a,b,c)}}{\partial{c}} = 0

We have the equations :

a 3 + ( a 2 c ) 3 ( a b ) 3 = b 3 + ( b 2 a ) 3 ( b c ) 3 = c 3 + ( c 2 b ) 3 ( c a ) 3 = a 3 + ( a 2 c ) 3 ( a b ) 3 + b 3 + ( b 2 a ) 3 ( b c ) 3 + c 3 + ( c 2 b ) 3 ( c a ) 3 3 \begin{aligned} a^3+(a^2c)^3 - (ab)^{-3} = b^3+(b^2a)^3 - (bc)^{-3} = c^3+(c^2b)^3 - (ca)^{-3} = \frac{a^3+(a^2c)^3 - (ab)^{-3} + b^3+(b^2a)^3 - (bc)^{-3} + c^3+(c^2b)^3 - (ca)^{-3}}{3}\end{aligned}

Since the combined equation is homogeneous in ( a , b , c ) (a,b,c) by WLOG we have a b c a\le b\le c & for minimum we must have a b c a + b + c 3 1 2 a\le b \le c \le \frac{a+b+c}{3} \le \frac{1}{2}

Putting the minimum value of a , b , c a,b,c in the expression we get maximum as 6.031 \boxed{6.031}

f ( a , b , c ) a , f ( a , b , c ) b , f ( a , b , c ) c = 0 \frac{\partial{f(a,b,c)}}{\partial{a}} , \frac{\partial{f(a,b,c)}}{\partial{b}},\frac{\partial{f(a,b,c)}}{\partial{c}} = 0

How do you know you've found the global maximum point? And not a local minimum point, saddle point, etc?

Pi Han Goh - 5 years, 2 months ago

Log in to reply

he can take the hessian then take the determinants of the principal minors

Hamza A - 5 years, 2 months ago

Log in to reply

If that's the case, he is missing out on a lot of steps. Plus, I'm sure it's not easy to find the partial derivatives for most of the expressions in the matrix.

Pi Han Goh - 5 years, 2 months ago

Log in to reply

@Pi Han Goh yeah,it'll take many steps,which he needs to show

Hamza A - 5 years, 2 months ago

The solution is not for the maxima, but for a minima. You can verify that this function is strictly a convex function which is asked to be maximized over a convex constraint set. Had the question asked for a minimization then the function would have a unique global minimizer in the interior of this convex constarint set, that is a + b + c 3 / 2 , a > 0 , b > 0 , c > 0 a+b+c\le 3/2,\ a> 0,\ b> 0,\ c> 0 . You've got this point as your answer, which is a = b = c = 1 / 2 a=b=c=1/2 . On the other hand, a convex function is maximized over a compact convex constraint set at its extreme points. Since the constraint set here is convex, but not compact (since it is not closed due to the strict poisitivity requirement of a , b , c a,b,c ), the function cannot be maximized, that is it can be made arbitrarily large by taking suitable a , b , c a,b,c .To verify my point, you can just take a = 3 / 2 2 × 1 0 4 = 1.4998 , b = c = 1 0 4 a=3/2-2\times 10^{-4}=1.4998,\ b=c=10^{-4} which are well within the constraint set, but gives the function a value 1 0 4 \approx 10^4 which is way larger than 6.031 6.031 . So the question is wrong and your answer finds, by mistake, the minimizer of the function (which the question should have asked).

Samrat Mukhopadhyay - 5 years, 2 months ago

Log in to reply

A modification has been made, sorry for all the inconvenience!

Leah Jurgens - 5 years, 2 months ago

Can someone plz explain me this sum?

Vrushali Behara, - 5 years, 2 months ago

I am having a bit of difficulty understanding. Could someone explain

Aayush Patni - 5 years, 2 months ago

Log in to reply

Don't bother. His solution is wrong.

Pi Han Goh - 5 years, 2 months ago

Nice proof......sharma ji

Sayandeep Ghosh - 5 years, 1 month ago
Jeremi Litarowicz
Apr 11, 2016

Relevant wiki: Lagrange Multipliers

It is easy to see with the Minkowski inequality that: f ( a , b , c ) = a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 ( a + b + c ) 3 + ( 1 a + 1 b + 1 c ) 3 3 = g ( a , b , c ) f(a,b,c)=\sqrt [ 3 ]{ { a }^{ 3 }+\frac { 1 }{ { b }^{ 3 } } } +\sqrt [ 3 ]{ b^{ 3 }+\frac { 1 }{ c^{ 3 } } } +\sqrt [ 3 ]{ c^{ 3 }+\frac { 1 }{ a^{ 3 } } } \ge \sqrt [ 3 ]{ { (a+b+c) }^{ 3 }+{ (\frac { 1 }{ { a } } +\frac { 1 }{ { b } } +\frac { 1 }{ c } ) }^{ 3 } } =g(a,b,c)

Taking the directional derivative of g ( a , b , c ) g(a,b,c) allong v = [ 3 1 / 2 3 1 / 2 3 1 / 2 ] \overrightarrow { v } =\begin{bmatrix} { 3 }^{ -1/2 } \\ { 3 }^{ -1/2 } \\ { 3 }^{ -1/2 } \end{bmatrix} , we get: i = 1 3 1 3 1 3 ( g ( a , b , c ) ) 2 [ 3 ( a + b + c ) 2 3 x i 2 ( 1 a + 1 b + 1 c ) 2 ] \sum _{ i=1 }^{ 3 }{ \frac { 1 }{ \sqrt { 3 } } \frac { 1 }{ 3{(g(a,b,c))}^{2} } \left[ 3{ (a+b+c) }^{ 2 }-\frac { 3 }{ { x }_{ i }^{ 2 } } { (\frac { 1 }{ { a } } +\frac { 1 }{ { b } } +\frac { 1 }{ c } ) }^{ 2 } \right] }

Where x 1 = a , x 2 = b , x 3 = c {x}_{1}=a, {x}_{2}=b, {x}_{3}=c . We want to show that this is negative for a + b + c 3 / 2 a+b+c \le 3/2 . Exploiting the fact that g ( a , b , c ) > 0 g(a,b,c)>0 , for our condition to hold we get: i = 1 3 x i < i = 1 3 ( 1 x i j = 1 3 1 x j ) = ( j = 1 3 1 x j ) 2 \sum _{ i=1 }^{ 3 }{ { x }_{ i } } <\sum _{ i=1 }^{ 3 }{ (\frac { 1 }{ { x }_{ i } } \sum _{ j=1 }^{ 3 }{ \frac { 1 }{ { x }_{ j } } ) } } ={ \left( \sum _{ j=1 }^{ 3 }{ \frac { 1 }{ { x }_{ j } } } \right) }^{ 2 }

Assuming i = 1 3 x i = y \sum _{ i=1 }^{ 3 }{ { x }_{ i } } =y , we get that the maximum value of the smallest x i {x}_{i} is y / 3 y/3 . This implies that ( j = 1 3 1 x j ) 2 > 1 x n 2 9 y 2 {\left( \sum_{ j=1 }^{ 3 }{ \frac { 1 }{ { x }_{ j } } } \right) }^{ 2 } > \frac { 1 }{ { x }_{ n }^{ 2 } } \ge \frac { 9 }{ { y }^{ 2 } } . Therefore our condition definitely holds for y < 9 y 2 y < 9 1 / 3 2.08 y<\frac { 9 }{ { y }^{ 2 } } \Rightarrow y<{9}^{1/3}\approx 2.08 .

From this we can conclude that the minimum of g ( a , b , c ) g(a,b,c) must lie on a + b + c = 3 / 2 a+b+c=3/2 We now create the lagrange multiplier of g ( a , b , c ) 3 {g(a,b,c)}^{3} , which has the same minimum as g ( a , b , c ) g(a,b,c) , with the constraint a + b + c = 3 / 2 a+b+c=3/2 . L ( a , b , c , λ ) = ( a + b + c ) 3 ( 1 a + 1 b + 1 c ) 3 λ ( a + b + c 3 / 2 ) = 27 / 8 ( 1 a + 1 b + 1 c ) 3 λ ( a + b + c 3 / 2 ) L (a,b,c,\lambda )={ (a+b+c) }^{ 3 }-{ (\frac { 1 }{ { a } } +\frac { 1 }{ { b } } +\frac { 1 }{ c } ) }^{ 3 }-\lambda (a+b+c-3/2)=27/8-{ (\frac { 1 }{ { a } } +\frac { 1 }{ { b } } +\frac { 1 }{ c } ) }^{ 3 }-\lambda (a+b+c-3/2) Taking the derivative with respect to x i {x}_{i} : x i ( L ( x 1 , x 2 , x 3 , λ ) ) = 3 ( 1 a + 1 b + 1 c ) 2 1 x i 2 λ { \partial }_{ { x }_{ i } }(L({ x }_{ 1 },{ x }_{ 2 },{ x }_{ 3 },\lambda))=3{ (\frac { 1 }{ { a } } +\frac { 1 }{ { b } } +\frac { 1 }{ c } ) }^{ 2 }\frac { 1 }{ { x }_{ i }^{ 2 } } -\lambda

Since x i ( L ( x 1 , x 2 , x 3 , λ ) ) = 0 { \partial }_{ { x }_{ i } }(L({ x }_{ 1 },{ x }_{ 2 },{ x }_{ 3 },\lambda))=0 at the minimum: 3 ( 1 a + 1 b + 1 c ) 2 1 x i 2 λ = 0 3{ (\frac { 1 }{ { a } } +\frac { 1 }{ { b } } +\frac { 1 }{ c } ) }^{ 2 }\frac { 1 }{ { x }_{ i }^{ 2 } } -\lambda =0 ( 1 a + 1 b + 1 c ) 3 / λ = x i { (\frac { 1 }{ { a } } +\frac { 1 }{ { b } } +\frac { 1 }{ c } ) }\sqrt { 3/\lambda } ={ x }_{ i } a = b = c = 1 / 2 a=b=c=1/2 Therefore the minimum value of g ( a , b , c ) g(a,b,c) is 1755 3 / 2 \sqrt [ 3 ]{ 1755 } /2 . If we evaluate f ( 1 / 2 , 1 / 2 , 1 / 2 ) f(1/2,1/2,1/2) , we get the same value.

Since g ( a , b , c ) f ( a , b , c ) g(a,b,c)\le f(a,b,c) , the minimum value of f ( a , b , c ) f(a,b,c) is 1755 3 / 2 6.031 \sqrt [ 3 ]{ 1755 } /2\approx 6.031 .

Alex Alex
Apr 16, 2016

Let

x = a + b + c , y = 1 a + 1 b + 1 c x = a + b + c, y = \frac{1}{a} + \frac{1}{b} + \frac{1}{c} .

Using Minkowski inequality we get:

a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 ( a + b + c ) 3 + ( 1 a + 1 b + 1 c ) 3 3 = x 3 + y 3 3 \sqrt[3]{ a^3 + \frac{1}{b^3} } + \sqrt[3]{ b^3 + \frac{1}{c^3} } + \sqrt[3]{ c^3 + \frac{1}{a^3} } \ge \sqrt[3]{ (a + b + c)^3 +\big( \frac{1}{a} + \frac{1}{b} +\frac{1}{c} \big)^3 } = \sqrt[3]{ x^3 +y^3 }

From the statement x 3 / 2 x \le 3/2 .

From Arithmetic-Harmonic Mean Inequality we get:

3 y x 3 1 / 2 \frac{3}{y} \le \frac{x}{3} \le {1/2}

From here we can find out that x y 9 xy \ge 9 and y 6 y \ge 6

Now let us use Cauchy inequality for 65 terms:

x 3 + 64 1 64 y 3 65 x 3 ( y 4 ) 64 3 65 = 65 ( x y 4 ) 3 ( y 4 ) 63 3 65 65 ( 9 4 ) 3 ( 6 4 ) 63 3 65 = 65 ( 3 2 ) 3 x^3 + 64 \cdot \frac{1}{64} y^3 \ge 65 \sqrt[65] { x^3 \cdot \big( \frac{y}{4} \big)^{64 \cdot 3} } = 65 \sqrt[65] { \big(\frac{xy}{4}\big)^3 \cdot \big( \frac{y}{4} \big)^{63 \cdot 3} } \ge 65 \sqrt[65] { \big(\frac{9}{4}\big)^3 \big( \frac{6}{4} \big)^{63 \cdot 3} } = 65 \cdot \big( \frac{3}{2} \big)^3

(It may be unclear why do I take exactly these terms. We want equality to be achieved in Cauchy inequality so all terms should be equal and since intuitively a = b = c = 1 / 2 a = b = c = 1/2 in minimum, then we should use inequality for the equal terms: x 3 , ( y / 4 ) 3 , ( y / 4 ) 3 , . . . x^3, (y/4)^3, (y/4)^3, ... . This is of course informal explanation.)

Now we finally get

a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 x 3 + y 3 3 3 2 65 3 6.031 \sqrt[3]{ a^3 + \frac{1}{b^3} } + \sqrt[3]{ b^3 + \frac{1}{c^3} } + \sqrt[3]{ c^3 + \frac{1}{a^3} } \ge \sqrt[3] { x^3 + y^3 } \ge \frac{3}{2} \sqrt[3] { 65} \approx \boxed{6.031}

From the Minkowski inequality we get that equality holds if and only if a = b = c = 1 2 a = b = c = \frac{1}{2} .

Son Nguyen
Apr 12, 2016

Set the polynomial is P By applying C-S consequence(Mincowski) inequality: P ( x + y + z ) 3 + ( 1 x + 1 y + 1 z ) 3 3 27 x y z + 1 27 x y z 3 P\geq \sqrt[3]{(x+y+z)^{3}+(\frac{1}{x}+\frac{1}{y}+\frac{1}{z})^3}\geq \sqrt[3]{27xyz+\frac{1}{27xyz}} Set x y z = t ; 0 < t 1 2 xyz=t;0< t\leq \frac{1}{2} We see f ( t ) f(t) is decreasing in that interval. So m i n f ( t ) = f ( 1 8 ) = 1755 8 min f(t)=f(\frac{1}{8})=\frac{1755}{8}

So P 1755 8 3 6.031 P\geq \sqrt[3]{\frac{1755}{8}}\approx 6.031 .

Your solution is very good and short! You have a typo you may want to fix though. It should be 27/(xyz) instead of 1/(27xyz). (He used x, y, z instead of a, b, c by the way.) Is the condition for two left inequalities to hold the fact that the vectors [a, 1/b], [b, 1/c], [c, 1/a] are all in proportion?

James Wilson - 3 years, 9 months ago
Jack Lam
Apr 10, 2016

By Minkowski's Inequality for p < 1 p<1 , we have

( a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 ) 3 ( a + b + c ) 3 + ( 1 a + 1 b + 1 c ) 3 \left(\sqrt[3]{a^3+\frac{1}{b^3}}+\sqrt[3]{b^3+\frac{1}{c^3}}+\sqrt[3]{c^3+\frac{1}{a^3}}\right)^3 \geq (a+b+c)^3 + \left(\frac{1}{a}+\frac{1}{b}+\frac{1}{c}\right)^3

By the Cauchy–Schwarz Inequality or Arithmetic-Harmonic Mean Inequality, it is easy to show that:

( 1 a + 1 b + 1 c ) 3 216 \left(\frac{1}{a}+\frac{1}{b}+\frac{1}{c}\right)^3 \geq 216

( a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 ) 3 ( a + b + c ) 3 + 216 \left(\sqrt[3]{a^3+\frac{1}{b^3}}+\sqrt[3]{b^3+\frac{1}{c^3}}+\sqrt[3]{c^3+\frac{1}{a^3}}\right)^3 \geq (a+b+c)^3 + 216

Note that this occurs only when a = b = c = 1 2 a=b=c=\frac{1}{2}

Also note that a + b + c a+b+c is maximised at the same values

( a 3 + 1 b 3 3 + b 3 + 1 c 3 3 + c 3 + 1 a 3 3 ) 3 27 8 + 216 \left(\sqrt[3]{a^3+\frac{1}{b^3}}+\sqrt[3]{b^3+\frac{1}{c^3}}+\sqrt[3]{c^3+\frac{1}{a^3}}\right)^3 \geq \frac{27}{8} + 216

Unfortunately, the above logic is not correct, and therefore, invalid, even though it is the true minimum of the expression. I presently do not see a viable means of justifying the above conclusion.

Discuss in the comments.

Leah Jurgens
Apr 10, 2016

Since a lot of people have complained about this asking for minimum in stead of maximum, I went to check very carefully and realized my mistake. Therefore, a (not at all) slight modification has been made. Sorry for all the inconvenience!

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...