Finding the smallest value satisfying an inequality

  • Thread starter Fellowroot
  • Start date
  • Tags Tags
    Value
6 replies · 2K views · last reply
Fellowroot
Messages
92
Reaction score
0

Homework Statement



Typeset problem: Find, with proof, the smallest possible value for 1 - 1/n - 1/m - 1/k, where k, m and n are three different positive integers and 1/n + 1/m + 1/k < 1.

Homework Equations



n/a

The Attempt at a Solution



Here is one attempt

Handwritten-style derivation of algebraic steps: y = 1 - (1/n) - (1/m) - (1/k), leading through squaring to 2(k + m + n) < knm.


But I'm stuck on this inequality. I can't go further.

and here is another, but I don't know if I proved anything here.

Handwritten-style derivation of inequalities: 1 - (1/n) - (1/m) - (1/k) = min (1); (1/n) + (1/m) + (1/k) = -min + 1 (2); (1/n) + (1/m) + (1/k) < 1 (3); plug (2) into (3); -min + 1 < 1; -min < 0; min > 0.


Really looking if anyone could help me on this or if I'm on the right track. Thanks.
 
There are two components to the problem; (1) find a candidate for the solution and (2) prove that candidate is the right one.

For (1) I recommend throwing away all this fancy schmancy algebra and calculus and just roll up your sleeves and try a few things. Also, recognizing that your problem is equivalent to maximizing ##\frac{1}{n}+\frac{1}{m}+\frac{1}{k}## subject to ##n,m,k## distinct and ##\frac{1}{n}+\frac{1}{m}+\frac{1}{k}<1## might make some of this work a little more manageable.

Once you've found a triplet that works, try to prove that it's the best triplet. Don't get fancy, just think about it. If need be, find other triplets that work (in the sense that ##n,m,k## are distinct and ##\frac{1}{n}+\frac{1}{m}+\frac{1}{k}<1##) and try to see why your triplet is better.
 
Fellowroot said:

Homework Statement



Math problem text: Find, with proof, the smallest possible value for 1 - 1/n - 1/m - 1/k, where k, m and n are three different positive integers and 1/n + 1/m + 1/k < 1.



Homework Equations



n/a



The Attempt at a Solution



Here is one attempt

Handwritten-style derivation of y = 1 - (1/n) - (1/m) - (1/k), with y' = n^-2 + m^-2 + k^-2, set to 0, then (1/n) + (1/m) + (1/k) < 1, squared to 1/n^2 + 2/(nm) + 2/(nk) + 2/(mk) + 1/m^2 + 1/k^2 < 1, reduced to (k+m+n)/(knm) < 1/2, and 2(k+m+n) < knm.


But I'm stuck on this inequality. I can't go further.

and here is another, but I don't know if I proved anything here.

Handwritten-style derivation with equations (1) to (3) and steps: 1 - (1/n) - (1/m) - (1/k) = min; (1/n) + (1/m) + (1/k) = -min + 1; then plugged into (1/n) + (1/m) + (1/k) < 1, giving -min + 1 < 1, -min < 0, min > 0.


Really looking if anyone could help me on this or if I'm on the right track. Thanks.

No, you are on the wrong track: you cannot take derivatives with respect to discrete (integer-valued) variables like n, m and k. Derivatives need continuous variables, and you don't have those in this problem.
 
And even if m, n, and k were continuous variables, you certainly cannot take the derivative with respect to three different variables as you did here.
 
Thanks for pointing that out because I just remembered that it would be an implicit differentiation if I did take the derivative and I did not do that.

recognizing that your problem is equivalent to maximizing 1/n+1/m+1/k

Thanks for the hint, but I don't quite understand it. I thought I was trying to minimize it not maximize it.
 
In this case if you maximize 1/n+1/m+1/k (while still being smaller than 1) you minimize your goal function, right?