Too generalized?

Algebra Level 5

Over all positive integers n n and for all n n numbers a 1 , a 2 , . . . , a n a_1,a_2,...,a_n satisfying 1 a 1 , a 2 , . . . , a n 1 and i = 1 n a i 3 = 0 , -1 \le a_1,a_2,...,a_n \le 1 \ \text{ and }\ \sum_{i=1}^{n} {a}_{i}^{3}=0, find the maximum value of i = 1 n a i n \displaystyle \frac{\sum_{i=1}^{n} {a}_{i}}{n} to 3 decimal places.


Inspired by this person .
Here is an easier version .


The answer is 0.333.

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.

2 solutions

Mark Hennings
Mar 3, 2018

@Jon Haussmann 's proof of the special case generalises easily. Note that 4 a j 3 3 a j + 1 = ( 2 a j 1 ) 2 ( a j + 1 ) 0 4a_j^3 - 3a_j + 1 \; = \; (2a_j - 1)^2(a_j+1) \; \ge \; 0 for all j j , so that 3 j = 1 n a j 4 j = 1 n a j 3 + n = n 3\sum_{j=1}^n a_j \; \le \; 4\sum_{j=1}^n a_j^3 + n \; = \; n and hence 1 n j = 1 n a j 1 3 \tfrac{1}{n}\sum_{j=1}^n a_j \;\le \; \tfrac13 This inequality will become an equality if all the a j a_j can be chosen to be equal to either 1 -1 or 1 2 \tfrac12 , which means that there exist non-negative integers a , b a,b such that a + b = n a+b = n and a + 1 8 b = 0 -a + \tfrac18b = 0 . This is possible if n n is divisible by 9 9 , with a = 1 9 n a= \tfrac19n and b = 8 9 n b= \tfrac89n . Thus, maximising over all n n , see that the upper bound is 1 3 \boxed{\tfrac13} .

This approach is known as the tangent line inequality , which is useful in cases where the variables are separable (no cross terms).

The motivation for the inequality in the first line is consider the tangent to y = x 3 y = x^3 , that passes through ( 1 , 1 ) (-1, -1) . The point of tangent occurs at x = 1 2 x = \frac{ 1}{2} . We then take x 3 x^3 - "equation for this tangent line" to get us a bound involving x 3 x^3 and x x .
Since we know that this is a cubic which passes through x = 1 x = -1 , and is tangential at x = 1 2 x = \frac{1}{2} , we can conclude that the relevant inequality is ( x 1 2 ) 2 ( x ( 1 ) ) 0 (x - \frac{1}{2} ) ^2 ( x - (-1) ) \geq 0 .

Calvin Lin Staff - 3 years, 3 months ago

Log in to reply

Thanks for the mentioning. I didn’t know this :)

Steven Jim - 3 years, 3 months ago

Great solution! Thanks!

Steven Jim - 3 years, 3 months ago

how about if n n is not divisible by 9?

I Gede Arya Raditya Parameswara - 3 years, 2 months ago

Log in to reply

If so, you can not pick non-negative integers a , b a,b satisfying a + b = n a+b=n and b 8 a = 0 b-8a=0 , thus the maximum in such cases can't be 1 / 3 1/3 .

Steven Jim - 3 years, 2 months ago

Log in to reply

What Steven meant is that this approach will not work.

I suspect there isn't an easy way of finding the maximum then (though we know it is strictly smaller then 1/3)

Calvin Lin Staff - 3 years, 2 months ago

We want to maximize f ( x ) = j = 1 n sin x j f(\mathbf{x}) \; = \; \sum_{j=1}^n \sin x_j subject to the constraint g ( x ) = j = 1 n sin 3 x j = 0 g(\mathbf{x}) \; = \; \sum_{j=1}^n \sin^3x_j \; = \; 0 For simplicity, suppose that n 9 n \ge 9 . Then the case sin x 1 = 1 \sin x_1 = -1 , sin x 2 = sin x 3 = = sin x 9 = 1 2 \sin x_2 = \sin x_3 = \cdots = \sin x_9 = \tfrac12 , with sin x j = 0 \sin x_j = 0 for j > 9 j > 9 , shows that the maximum possible value of f f subject to this constraint will be positive.

Since there is no boundary to the domain for x \mathbf{x} , the maximum of f f subject to the constraint g = 0 g=0 will be identified by the method of Lagrange multipliers, so there exists λ \lambda such that f λ g = 0 \nabla f - \lambda \nabla g = 0 , and hence cos x j ( 1 3 λ sin 2 x j ) = 0 1 j n \cos x_j\big(1 - 3\lambda \sin^2x_j\big) \; = \; 0 \hspace{2cm} 1 \le j \le n If cos x j = 0 \cos x_j = 0 for all j j , then sin x j = ± 1 \sin x_j = \pm1 for all j j , which would imply that f = 0 f = 0 . Thus, for a positive value of f f , there must be a number 0 < μ < 1 0 < \mu < 1 such that sin x j { 1 , 1 , μ , μ } \sin x_j \in \{1,-1,\mu,-\mu\} for all j j . Thus there must exist nonnegative integers a , b , c , d a,b,c,d such that a + b + c + d = n a b + μ 3 ( c d ) = 0 a b + μ ( c d ) = f \begin{aligned} a + b + c + d & = \; n \\ a - b + \mu^3(c - d) & = \; 0 \\ a - b + \mu(c - d) & = \; f \end{aligned} (so that a a of the sin x j \sin x_j are equal to 1 1 , b b are equal to 1 -1 , c c are equal to μ \mu and d d are equal to μ -\mu ). Since f > 0 f > 0 we deduce that c > d c > d , and hence that b > a b > a . Moreover b a < c d b - a < c - d , and we have μ = ( b a c d ) 1 3 f = ( b a ) 1 3 ( c d ) 2 3 ( b a ) \mu \; = \; \left(\frac{b-a}{c-d}\right)^{\frac13} \hspace{2cm} f \; = \; (b-a)^{\frac13}(c-d)^{\frac23} - (b-a) For a given choice of a , b a,b , it is clear that f f will be maximized by maximizing c d c-d , so we must have c = n a b c = n-a-b and d = 0 d=0 , and hence f = ( b a ) 1 3 ( n a b ) 2 3 ( b a ) f \; = \; (b - a)^{\frac13}(n - a - b)^{\frac23} - (b-a) Putting t = b a t = b - a , we see that 0 t n 2 a 0 \le t \le n-2a and that f = t 1 3 ( n 2 a t ) 2 3 t f \; = \; t^{\frac13}(n-2a-t)^{\frac23} - t From this it is clear that, for a given choice of t t , f f is maximized by minimizing a a , and so a = 0 a=0 . Thus we are picking a = d = 0 a=d=0 and b + c = n b + c = n where b < c b < c , and so we want to maximize f = b 1 3 ( n b ) 2 3 b 1 b < 1 2 n f \; = \; b^{\frac13}(n-b)^{\frac23} - b \hspace{2cm} 1 \le b < \tfrac12n The function F ( u ) = u 1 3 ( 1 u ) 2 3 u 0 < u < 1 F(u) \; = \; u^{\frac13}(1-u)^{\frac23} - u \hspace{2cm} 0 < u < 1 has its maximum value of 1 3 \tfrac13 at u = 1 9 u = \tfrac19 , and is increasing for u < 1 9 u < \tfrac19 and decreasing for u > 1 9 u > \tfrac19 . Thus we deduce that f f will be equal to either F ( 1 9 n ) F\big(\lfloor \tfrac19n\rfloor \big) or else F ( 1 9 n + 1 ) F\big(\lfloor \tfrac19n\rfloor +1\big) , whichever is larger.

Thus the maximum value of n 1 j = 1 n y j n^{-1}\sum_{j=1}^ny_j subject to the constraint j = 1 n y j 3 = 0 \sum_{j=1}^n y_j^3 = 0 and y j 1 |y_j| \le 1 for all 1 j n 1 \le j \le n is 1 n [ b 1 3 ( n b ) 2 3 b ] \tfrac{1}{n}\big[ b^{\frac13}(n-b)^{\frac23} - b \big] where b b is either 1 9 n \lfloor \tfrac19n\rfloor or 1 9 n + 1 \lfloor \tfrac19n\rfloor + 1 , and this value is achieved by choosing b b of the y j y_j equal to 1 -1 , with the remaining n b n-b values of y j y_j equal to ( b n b ) 1 3 \big(\tfrac{b}{n-b}\big)^{\frac13} . This maximum value is less than or equal to 1 3 \tfrac13 , being equal to 1 3 \tfrac13 when 9 9 divides n n . I have not investigated which of the two possible values of b b need to be chosen for any other particular value of n n , but in many cases we choose b b to the integer closest to 1 9 n \tfrac19n .

@Calvin Lin

Mark Hennings - 3 years, 2 months ago
Ivo Zerkov
Mar 2, 2018

This is a constrained optimization problem, with a Lagrangian of L ( a 1 , a 2 , , a n ) = a 1 + a 2 + a n λ ( a 1 3 + a 2 3 + + a n 3 ) \large \mathcal{L}(a_{1},a_2,\cdots,a_{n})=a_{1}+a_{2}\dots+a_{n}-\lambda\cdot(a_{1}^3+a_{2}^3+\dots+a_{n}^3) . Setting L = 0 \large \nabla\mathcal{L}=0 yields a 1 = ± a 2 = ± a 3 = ± a n a_1=\pm a_2=\pm a_3=\dots\pm a_n . It's clearly unoptimal for that to be true, since f = a 1 + a 2 + a n n = 0 f=\frac{a_1+a_2+\dots a_n}{n}=0 .

Hence at least one of the values is on the boundary, namely a n = ± 1 a_n=\pm 1 . Considering the new function with its new constraints, we again have a 1 = ± a 2 = = ± a n 1 a_1=\pm a_2=\dots=\pm a_{n-1} . Since having a k = a j a_k=-a_j here makes no impact on the final sum but enlarges n n , they must either all be equal or another one has to be on the boundary. Continuing in this fashion, we have x x of the n n numbers on the (same) boundary, and the rest equal. Since differentiating between the 1 -1 and 1 1 cases for the boundary simply changes f f 's sign, we assume WLOG a 1 = a 2 = = a x = 1 a_1=a_2=\dots=a_x=-1 .

Thus a x + 1 = a x + 2 = = a n = ( x n x ) 1 3 a_{x+1}=a_{x+2}=\dots =a_n=(\frac{x}{n-x})^{ \frac{1}{3}} , and so f = ( n x ) 2 3 x 1 3 x n f=\frac{(n-x)^{\frac{2}{3}}\cdot x^{\frac{1}{3}}-x}{n} . Letting x 1 3 = a x^{\frac{1}{3}}=a and ( n x ) 1 3 = k a (n-x)^{\frac{1}{3}}=k\cdot a with k 1 k\geq 1 , which is what our boundary condition a p 1 a_p\leq1 implies, we have f = k 2 a 3 a 3 a 3 ( k 3 + 1 ) = k 1 k 2 k + 1 f=\frac{k^2\cdot a^3-a^3}{a^3\cdot(k^3+1)}=\frac{k-1}{k^2-k+1} . This reaches its maximum for k = 2 k=2 , hence n x = k 3 x = 8 x n-x=k^3\cdot x=8\cdot x , making n = 9 x n=9\cdot x and f = k 1 k 2 k + 1 = 1 3 f=\frac{k-1}{k^2-k+1}=\frac{1}{3} , which is the answer.

Is there a non-Lagrangian solution for this problem?

Good solution anyways!

Steven Jim - 3 years, 3 months ago

Log in to reply

Probably, seeing as I only really use the Lagrangian to show that the a a 's must either be on the boundary or equal to each other. I think an argument can be made about there being an "optimal" value for the a a 's not on the boundary, hence why they're all equal. Either way, I think using the Lagrangian is the easiest way to justify the problem rigorously.

Ivo Zerkov - 3 years, 3 months ago

The maximum of f ( x ) = 1 n [ x 1 3 ( n x ) 2 3 x ] f(x) = \tfrac{1}{n}\big[x^{\frac13}(n-x)^{\frac23} - x\big] over 0 x 1 2 n 0 \le x \le \tfrac12n is indeed 1 3 \tfrac13 , achieved when x = 1 9 n x = \tfrac19n .

We need a slight nuance to your proof. If n n is not a multiple of 9 9 , then the above maximum of f f is not achieved at an integer value of x x , and so the maximum value of 1 n j = 1 n a j \tfrac{1}{n}\sum_{j=1}^n a_j , subject to the given constraints, is less than 1 3 \tfrac13 . For example, when n = 10 n=10 the maximum is 1 10 ( 3 4 3 1 ) \tfrac{1}{10}\big(3^{\frac43} - 1\big) . However, the maximum value is exactly 1 3 \tfrac13 whenever n n is a multiple of 9 9 , and so (since the question also asks us to maximize f f over n n ) the answer of 1 3 \tfrac13 is correct.

Mark Hennings - 3 years, 3 months ago

Log in to reply

Yes, that is worth mentioning.

Ivo Zerkov - 3 years, 3 months ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...