As x , y , z ranges over all possible real numbers, what is the minimum value of
3 x 2 + 1 2 y 2 + 2 7 z 2 − 4 x y − 1 2 y z − 6 x z − 8 y − 2 4 z + 1 0 0 ?
This problem has been proposed by Hui .
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.
Nice solution but how did you find your sum of squares expression for F?
Log in to reply
I have another way to solve this problem by taking derivation of F with respect to x,y,z and get x,y,z by solving equations. However, with this method, it is hard to explain whether F reach its minimum or maximum. But with the result of x,y,z, I can easily rewrite F as the sum of squares.
Can you describe the thought process in factoring the original polynomial into the form you found?
Log in to reply
Yes, as I mentioned above, I took the derivation of F and get x = 2 3 , y = 1 , z = 6 5 . Therefore, the sum of squares must contain ( 2 x − 3 y ) 2 , ( 6 y − 5 z ) 2 , ( 9 z − 5 x ) 2 , ( 2 x − 3 ) 2 , ( y − 1 ) 2 , ( 6 z − 5 ) 2 . Then, from the original function, I found the coefficient of these squares, which are 3 1 , 5 1 , 1 5 1 , 0 , 4 , 5 2 respectively.
Log in to reply
Ah okay. I thought you had found a way to factor with no derivative takin.
Log in to reply
@Josh Silverman – I tried to make some squares and got (x-3z)²+2(x-y)²+(2y-3z)²+6(y-2/3)²+9(z-4/3)²+244/3. But I am unable to get the minimum value as I think I can't simply say it is 244/3 cuz all the other terms are not necessarily 0 if I put some values of x,y,z to eliminate some of the terms. All the terms cannot get eliminated. So I did a little bit of case work but failed to find an integral value. Where did I go wrong? Is my initial expression wrong? Am I missing something?
@Josh Silverman – Is there any way algebraic strategy to factorize this expression that doesn't employ calculus?
The function is a polynomial of degree 2 , a quadric surface in R 3 : f ( x , y , z ) = 3 x 2 + 1 2 y 2 + 2 7 z 2 − 4 x y − 1 2 y z − 6 x z − 8 y − 2 4 z + 1 0 0 ; the minimum is in a stationary point P ∗ , so it must be searched in the solution of:
g r a d ( f ) = 0 ; which leads to:
∂ x i ∂ f = 0 ; i = 1 , 2 , 3 ;
x 1 = x ; x 2 = y ; x 3 = z ; so:
⎩ ⎪ ⎨ ⎪ ⎧ ∂ x ∂ f = 6 x − 4 y − 6 z = 0 ∂ y ∂ f = 2 4 y − 4 x − 1 2 z − 8 = 0 ∂ z ∂ f = 5 4 z − 1 2 y − 6 x − 2 4 = 0
symplifing, it can be obtained:
⎩ ⎪ ⎨ ⎪ ⎧ 3 x − 2 y − 3 z = 0 x − 6 y + 3 z = − 8 x + 2 y − z = − 4
this is a system of 3 linear equations in the variables x , y , z and admit an unique solution:
P ∗ = ( x ∗ , y ∗ , z ∗ ) ;
x ∗ = 2 3 , y ∗ = 1 , z ∗ = 6 5 ,
f ( P ∗ ) = 8 6 ;
and it is the minimum, because the Hessian matrix in P ∗ ,
H ( f ( P ) ) ∣ P ∗ = { ∂ x i ∂ x j ∂ f 2 } ∣ P ∗ ,
is definite positive:
H ( f ( P ) ) ∣ P ∗ = 4 8 × ⎩ ⎨ ⎧ 3 − 1 − 1 − 2 6 − 2 − 3 − 3 9 ⎭ ⎬ ⎫ ,
because the north - western matrix determinant minors are:
M 1 × 1 = 4 8 × 3 > 0 ;
M 2 × 2 = 4 8 × 1 6 > 0 ;
M 3 × 3 = d e t H ( f ( P ) ) ∣ P ∗ = 4 8 × 2 1 0 > 0 ;
so, by Sylvester's criteria the eigenvalues of the Hessian matrix are positive, the Hessian matrix evaluated in P ∗ and the squared form associated S ( X ) are definite positive:
S ( X ) = X T H X > 0 , ∀ X = 0 ,
so by the theory of function of several variables, P ∗ is a point of minima.
I tried factorizing but it was hopeless, then I had to use calculus to solve this one, but isn't there any other way to solve this question.
I tried to find solution without using calculus with integral coefficients but I failed to find it..
Log in to reply
I solved this without calculus and it's pretty straight forward(although it could take quiet a long time). Since there's no restrictions on x,y,z, I guess we are free to use the minimal property of a quadratic.
To see what I mean, consider the entire expression as a function of x, and assume that y,z are decided. Since this function is a quadratic that opens up, the minimum point occurs at the vertex(which is x = − b / 2 a for a x 2 + b x + c . This means that we can get x in terms of y,z if we just want the minimum. We plug x back in and we are left with a 2-variables quadratic. We repeat the same process to get y in terms of z and then plug in back in and now we just have a 1 variable quadratic quadratic which we know how to find the minimum of(using the same process).
I noticed that if we apply substitution a=2y and b=3z the whole thing seems simplified?
How did you rewrite the function in first step?
Let f ( x , y , z ) = 3 x 2 + 1 2 y 2 + 2 7 z 2 − 4 x y − 1 2 y z − 6 x z − 8 y − 2 4 z + 1 0 0 We are given that all variables are real numbers and aren't restricted in any way. Hence consider the system ∂ x ∂ f = 6 x − 4 y − 6 z ∂ y ∂ f = − 4 x + 2 4 y − 1 2 z − 8 ∂ z ∂ f = − 6 x − 1 2 y + 5 4 z − 2 4 This system has solution ( x , y , z ) = ( 2 3 , 1 , 6 5 ) . Hence the minimum is f ( 2 3 , 1 , 6 5 ) = 8 6 The fact that this is not maximum is obvious, the function f ( x , y , z ) approaches infinity for x approaching infinity and y , z zero. It could also be saddle point, but I won't discuss this here.
Call me slow, but can this system be used for any extrema or just this particular problem?
Very thorough and well explained. Love it
By any chance, is it possible to convert the given polynomial as ( a + b + c ) 2 ? to make things more simpler? Then we can make x and y terms 0 to get a constant remaining...that would be the minimum value.
This method is called "Lagrange Multiplier". A method to use in finding the maxima and minima of a multivariable function
Train of thoughts: Basically I attempted to factorize , although the expression did not look symmetric, which came out as nothing nicer than: x ( 3 x − 4 y − 6 z ) + y ( 1 2 y − 1 2 z − 8 ) + z ( 2 7 z − 2 4 ) + 1 0 0 . Maybe I missed something, but I gave up and went on to consider the polynomial discriminant , still in an attempt to learn more about the roots, which was obviously useless. Perhaps, I'm missing something trivial, but when one sees ugly powers and one wants to bring down the exponents , one way to do so is to consider calculus, i.e. derivatives . And clearly this expression when plotted is a concave up parabola, so calculus indeed works.
We first consider the derivative of the expression and set it to 0 to consider the relationship between x , y , z . Computing the derivative, we have:
d x d 3 x 2 + 1 2 y 2 + 2 7 z 2 − 4 x y − 1 2 y z − 6 x z − 8 y − 2 4 z + 1 0 0
= 3 d x d ( x 2 ) − 4 y d x d ( x ) − 6 z d x d ( x ) + d x d ( 1 2 y 2 ) + d x d ( 2 7 z 2 ) + d t d ( − 1 2 y z ) + d t d ( − 8 y ) + d t d ( − 2 4 z ) + d x d ( 1 0 0 ) (where we've differentiated each term and pulled out the constants)
= 6 x − 4 y − 6 z (recall that d t d ( y ) = 0 and d t d ( z ) = 0, etc. and the power rule)
Which we set to 0 , subbing 6 x = 4 y − 6 z back into the big expression, we get:
4 ( 6 x 2 − 6 x ( 2 y + 1 ) + 8 y 2 + 2 y + 2 5 ) ( ∗ )
That's the beauty of taking derivatives, we eliminate the variables one by one . Once again, and this time for simplicity sake, we take the derivative of 6 x 2 − 6 x ( 2 y + 1 ) + 8 y 2 + 2 y + 2 5 ( ∗ ∗ ) and set it to 0 :
d t d ( 6 x 2 − 6 x ( 2 y + 1 ) + 8 y 2 + 2 y + 2 5 ) = 6 d t d ( x 2 ) − 6 ( 2 y + 1 ) d t d ( x ) + d t d ( 8 y 2 ) + d t d ( 2 y ) + d t d ( 2 5 )
= 1 2 x − 6 ( 1 + 2 y ) = 6 ( 2 x − 2 y − 1 )
Now, 6 ( 2 x − 2 y − 1 ) = 0 ⇔ 2 x = 2 y + 1 , subbing this into ( ∗ ∗ ) , we get:
2 y 2 − 4 y + 2 4 7 .
With one variable, it is simple to complete the square , which then becomes: 2 ( y − 1 ) 2 + 2 4 3 . Clearly the minimum is 2 4 3 . Recall that in ( ∗ ) , we have to times 4 , so the final answer is 4 × 2 4 3 , which evaluates as 86 .
Oops I'm sorry there's some typos in the solution, please replace all d t d with d x d
Log in to reply
But how can you be sure it is minima? I mean, why cannot it be maxima?
Log in to reply
If 2nd differentiation is < or > o then the function is maxima or minima respectively
I've tackled a similar problem, but in that problem they completed the square in terms of ( a x + b y + c z + d ) 2 . I too have used that solution. But there must be a solution to this problem that does not make use of Calculus. I will try, though.
We will prove that the minimum value is 86. Applying AM-GM inequality yields: 3 4 x 2 + 3 y 2 ≥ 4 x y And 5 y 2 + 5 3 6 z 2 ≥ 1 2 y z And 5 2 7 z 2 + 3 5 x 2 ≥ 6 z x We also have: 4 ( y − 1 ) 2 ≥ 0 And 5 2 ( 6 z − 5 ) 2 ≥ 0 In conclusion, we have the minimum of the function is 86.
This is a sketch of a very nice solution. One plugs in the three listed inequalities to replace -4xy, -12yz and -6zx, and then completes a square. It is implicit in the solution that all inequalities become equalities for x=3/2, y=1, z=5/6.
Many solutions used Calculus technique: the minimum of a differentiable function of several variables can only occur when all partial derivatives are equal to zero. It is easy to use this to find a correct answer to this problem. However, it is not obvious that 86 is a global minimum (unless you assume that a minimum exists). One way to do it is by showing that the function is concave up everywhere (the matrix of second derivatives is positive-definite). Justifying this all rigorously is definitely much harder than completing the square or other similar technique. One can blend these approaches by first finding the minimum of f(x,y,z) for arbitrary fixed y and z, then finding the minimum of the resulting function of y and z for arbitrary fixed z and then finding the minimum of the resulting expression as a function of z. At every step, one only needs to minimize a quadratic function, with coefficients depending on some parameters.
Arrange the given expression, we get
2 ( x − y ) 2 + ( x − 3 z ) 2 + 6 ( y − z ) 2 + 4 ( y − 1 ) 2 + 1 2 ( z − 1 ) 2 + 8 4
Take ( y , z ) = ( 1 , 1 ) , because we will get three terms of 0 .
The two terms left, let it be A , are
2 ( x − y ) 2 + ( x − 3 z ) 2
= 3 x 2 + ( 4 y + 6 z ) x + ( 2 y 2 + 9 z 2 )
= 3 ( x − 3 2 y + 3 z ) 2 + ( 3 2 y 2 − 4 y z + 6 z 2 )
The minimum value of A is
3 2 y 2 − 4 y z + 6 z 2 .
When ( y , z ) = ( 1 , 1 ) , the value of A is 2 3 2 , the minimum value of the expression is 8 6 3 2 .
If we want to reduce the minimum value of A , we can either reduce y or z
By analysing, reduce the value of z is a better approach, so let y = 1 , z is reduced,
The value of A will be reduced by 2 − ( 6 z 2 − 4 z ) ,
The increase of the other terms will be 1 8 ( z − 1 ) 2 ,
The overall reduced value will be 2 − ( 6 z 2 − 4 z ) − 1 8 ( z − 1 ) 2 .
Now, the maximum value that can be reduced,
max [ 2 − ( 6 z 2 − 4 z ) − 1 8 z y − 1 ) 2 ] = 3 2
Hence, the minimum value of the expression is 8 6 3 2 − 3 2 = 8 6
The solution should be longer, since I skipped how I manage to find the arranged expression. Other than that, I think looking for the minimum value of A is the tedious part. If you have any better solution, please tell me...
Your solution isn't complete, since you only found the minimum when y = 1 . You will need to do this for all possible y values.
One approach would be to write the expression as the sum of squares. It helps to know that the minimum is achieved when x = 2 3 , y = 1 and z = 6 5 .
@Hui Xiong Care to comment?
Log in to reply
Thanks for the comment! Yes, I found where the mistake was when I was writing the solution so I asked for help in the comment section. This was the only way I found to express it as a sum of squares, do we have another one? Hope you can give me more hints and tips and I don't think @Hui Xiong is active recently but I hope he will check his email and give me some comments on this.
To get the minimum value of a function in x, we find the derivative of the function, equate it to 0 and solve for x, then substitute the answer to the function. The extension of this concept to a function of three variables (in x, y and z) entails us to find the partial derivative of the expression with respect to x, y and z, equate all of them to 0, then solve the resulting system of equations. The answers are then substituted to the original expression.
Let f ( x , y , z ) = 3 x 2 + 1 2 y 2 + 2 7 z 2 − 4 x y − 1 2 y z − 6 x z − 8 y − 2 4 z + 1 0 0 .
Then f x ( x , y , z ) = 6 x − 4 y − 6 z .
We also have f y ( x , y , z ) = 2 4 y − 4 x − 1 2 z − 8 .
Finally f z ( x , y , z ) = 5 4 z − 1 2 y − 6 x − 2 4 .
Now we equate each of them to 0. We have the ff. system of equations:
6 x − 4 y − 6 z = 0 ------------ (1)
− 4 x + 2 4 y − 1 2 z = 8 ------------ (2)
− 6 x − 1 2 y + 5 4 z = 2 4 ------------ (3)
We add (1) to (3) to get
− 1 6 y + 4 8 z = 2 4 ------------ (4)
We multiply both sides of (2) by 1 . 5 to get
− 6 x + 3 6 y − 1 8 z = 1 2 ------------ (5)
Now we add (1) to (5) to get
3 2 y − 2 4 z = 1 2 ------------ (6)
Now we divide both sides of (6) by 2 to get
1 6 y − 1 2 z = 6 ------------ (7)
We add (4) to (7) to get 36z = 30). Then \(z = \frac {5}{6} .
Now we substitute the value that we got for z to equation (7) to get
1 6 y − 1 2 ( 6 5 ) = 6 .
Solving for y, we have 1 6 y − 1 0 = 6 , 1 6 y = 1 6 and y = 1 . Finally we substitute the known values of y and z to equation (1) to get
6x - 4(1) - 6(\frac {5}(6}) = 0 .
Solving for x, we get 6 x − 4 − 5 = 0 , 6 x − 9 = 0 and x = 2 3 .
Now we have the values of x, y and z, we substitute each to the original expression. We have:
3 x 2 + 1 2 y 2 + 2 7 z 2 − 4 x y − 1 2 y z − 6 x z − 8 y − 2 4 z + 1 0 0
= 3 ( 2 3 ) 2 + 1 2 ( 1 ) 2 + 2 7 ( 6 5 ) 2 − 4 ( 2 3 ) ( 1 ) − 1 2 ( 1 ) ( 6 5 ) − 6 ( 2 3 ) ( 6 5 ) − 8 ( 1 ) − 2 4 ( 6 5 ) + 1 0 0
= 3 ( 4 9 ) + 1 2 + 2 7 ( 3 6 2 5 ) − 6 − 1 0 − 2 1 5 − 8 − 2 0 + 1 0 0
= 4 2 7 + 6 8 + 4 7 5 − 4 3 0
= 4 7 2 + 6 8
= 1 8 + 6 8
= 8 6
let f(x,y,z) equal the polynomial above, take partial derivatives by x, y and z to obtain the following equations
6x - 4y - 6z = 0 24y - 4x - 12z = 8 54z - 12y - 6x = 24
solve for x,y,z and check that it is a minimum
For single variable functions such as f ( x ) , we get the maximum/minimum by getting the first order derivative and equating it to zero. In cases such as this, we have multiple variables involved. f ( x , y , z ) can be partially differentiated by considering y and z as constants.
∂ x ∂ f ( x , y , z ) = ∂ x ∂ ( 3 x 2 − 4 x y − 6 x z )
To shorten the process, I have removed all terms independent of x { 1 2 y 2 , 2 7 z 2 , − 1 2 y z , − 8 y , − 2 4 z , 1 0 0 }. We find the first partial derivative to be 6 x − 4 y − 6 z . Notice that this is incomplete for we can only get the minimum of x if we have values for y and z , also at their minimum. Thus, we must get the partial derivatives in terms of y and z as well.
∂ y ∂ f ( x , y , z ) = ∂ y ∂ ( 1 2 y 2 − 4 x y − 1 2 y z − 8 y ) ∂ y ∂ f ( x , y , z ) = 2 4 y − 4 x − 1 2 z − 8
Again, we disregard all terms invariable with y , this time holding x and z constant. Repeating this process, we get the derivative of f ( x , y , z ) in terms of z . We get 5 4 z − 1 2 y − 6 x − 2 4 .
Equating everything to 0 , we get the minimum functions for x , y and z , respectively.
We can then proceed to solve for each. The solution is ( 2 3 , 1 , 6 5 ) - the minimum values for x, y and z. We can then use these to find the minimum value for the original function to get 8 6 .
The expression in the question can be rewritten as (2x-3y)^2/3+(5x-9z)^2/15+(5y-6z)^2/15+4(y-1)^2+2(6z-5)^2/5+86. So the minimum must be 86 achieved at (3/2, 1, 5/6).
define f as a function in x,y,z. solve the linear system: df/dx=0 df/dy=0 df/dz=0
f(x,y,z)=86
We differentiate the given expression with respect to x , y , z and get 6 x − 4 y − 6 z , − 4 x + 2 4 y − 1 2 z − 8 , − 6 x − 1 2 y + 5 4 z − 2 4 respectively. It is easy to see all three derivatives have their second-order derivative positive, which means that there is a global minimum. So we let 6 x − 4 y − 6 z = 0 , − 4 x + 2 4 y − 1 2 z − 8 = 0 , − 6 x − 1 2 y + 5 4 z − 2 4 = 0 , to find where the minimum is. Adding the first and third equalities gives 6 z − 2 y = 3 . Multiply the first by 2 and the second by 3 then add them up gives us 8 y − 6 z = 3 . Adding up these 2 equations give y = 1 . Substituting this into other equation gives us the value of x , z = 2 3 , 6 5 and locate the global minimum. Substituting these to the original expression gives 86.
Do partial differentiation of the equation to obtain 3 equations as follow and equate them them with 0,
3 x − 2 y − 2 z = 0 (simplified) − x + 6 y − 3 z − 2 = 0 − x − 2 y + 9 z − 4 = 0
Solving this we get the equation to be minima ( x , y , z ) = ( 2 3 , 1 , 6 5 )
Substituting this into original equation we get minima to be 86.
The analytic solution:
Define the following function: f = 3 x^2 + 12 y^2 + 27 z^2 - 4 x y - 12 y z - 6 x z - 8 y - 24*z + 100 Then the minimum occurs for solving ∇ f = 0 : Solve[{D[f, x] == 0, D[f, y] == 0, D[f, z] == 0}]
Of course one could derive and solve the system by hand, but since this is trivial, it its left out.
The numerical solution: FindMinimum[{3 x^2 + 12 y^2 + 27 z^2 - 4 x y - 12 y z - 6 x z - 8 y - 24*z + 100}, {{x, 1}, {y, 1}, {z, 1}}]
Problem Loading...
Note Loading...
Set Loading...
We can rewrite the given function by:
F = 3 1 ( 2 x − 3 y ) 2 + 5 1 ( 5 y − 6 z ) 2 + 1 5 1 ( 9 z − 5 x ) 2
+ 4 ( y − 1 ) 2 + 5 2 ( 6 z − 5 ) 2 + 8 6 .
It is clear that F ≥ 0 . F = 0 when 2 x − 3 y = 5 y − 6 z = 9 z − 5 x = y − 1 = 6 z − 5 = 0 .
These equations lead to x = 2 3 , y = 1 , z = 6 5 .
Therefore, the minimum value of F is 86, where x = 2 3 , y = 1 , z = 6 5 .