Thread Subject: optimization

Hi,

I have an objective function f(x) that is a matrix, this objective function is evaluated at different point in X(x,y) space

hence each Xi as a value f(Xi) associated with it, this f(Xi) is a function of some input vector [ui vi wi zi] which is bounded. The value of f(Xi) is influenced by all the i i=1...n
the f(X1)=f(X1,X2,....Xn)

I want f(Xi)=0 or close to zero for i=1...n by varying the [ui vi wi zi] vectors

can I used the simulated annealing tool box to perform this optimization, if not which matlab optimization tool can perform this analysis

Thanks

Annick

"Annick " <annick219@yahoo.ca> wrote in message <hcccum$idj$1@fred.mathworks.com>...
> Hi,
>
> I have an objective function f(x) that is a matrix, this objective function is evaluated at different point in X(x,y) space
>
> hence each Xi as a value f(Xi) associated with it, this f(Xi) is a function of some input vector [ui vi wi zi] which is bounded. The value of f(Xi) is influenced by all the i i=1...n
> the f(X1)=f(X1,X2,....Xn)
>
> I want f(Xi)=0 or close to zero for i=1...n by varying the [ui vi wi zi] vectors
>
> can I used the simulated annealing tool box to perform this optimization, if not which matlab optimization tool can perform this analysis
>
> Thanks
>
> Annick

You can use also ga or fmincon if I've understood correctly your problem.

"Luca Zanotti Fragonara" <Luca_Zanotti@libero.it> wrote in message <hccdhe$paf$1@fred.mathworks.com>...
> "Annick " <annick219@yahoo.ca> wrote in message <hcccum$idj$1@fred.mathworks.com>...
> > Hi,
> >
> > I have an objective function f(x) that is a matrix, this objective function is evaluated at different point in X(x,y) space
> >
> > hence each Xi as a value f(Xi) associated with it, this f(Xi) is a function of some input vector [ui vi wi zi] which is bounded. The value of f(Xi) is influenced by all the i i=1...n
> > the f(X1)=f(X1,X2,....Xn)
> >
> > I want f(Xi)=0 or close to zero for i=1...n by varying the [ui vi wi zi] vectors
> >
> > can I used the simulated annealing tool box to perform this optimization, if not which matlab optimization tool can perform this analysis
> >
> > Thanks
> >
> > Annick
>
> You can use also ga or fmincon if I've understood correctly your problem.

I am not sure that I can use fmincon because the minimization gives only one value

my
f as several values ex f=[2 3 1; 4 5 5; 7 1 3] and I want all the cells of f to be zero or close to zero

the value of f is obtain from a function that calculates the f values as a function of the input vector [u v w z]

which must be modified to define f=[0 0 0; 0 0 0; 0 0 0]

Thank's

"Annick " <annick219@yahoo.ca> wrote in message
news:hccfj2$4od$1@fred.mathworks.com...
> "Luca Zanotti Fragonara" <Luca_Zanotti@libero.it> wrote in message
> <hccdhe$paf$1@fred.mathworks.com>...
>> "Annick " <annick219@yahoo.ca> wrote in message
>> <hcccum$idj$1@fred.mathworks.com>...
>> > Hi,
>> >
>> > I have an objective function f(x) that is a matrix, this objective
>> > function is evaluated at different point in X(x,y) space
>> >
>> > hence each Xi as a value f(Xi) associated with it, this f(Xi) is a
>> > function of some input vector [ui vi wi zi] which is bounded. The value
>> > of f(Xi) is influenced by all the i i=1...n
>> > the f(X1)=f(X1,X2,....Xn)
>> >
>> > I want f(Xi)=0 or close to zero for i=1...n by varying the [ui vi wi
>> > zi] vectors
>> >
>> > can I used the simulated annealing tool box to perform this
>> > optimization, if not which matlab optimization tool can perform this
>> > analysis
>> >
>> > Thanks
>> >
>> > Annick
>>
>> You can use also ga or fmincon if I've understood correctly your problem.
>
> I am not sure that I can use fmincon because the minimization gives only
> one value
>
> my
> f as several values ex f=[2 3 1; 4 5 5; 7 1 3] and I want all the cells of
> f to be zero or close to zero
>
> the value of f is obtain from a function that calculates the f values as a
> function of the input vector [u v w z]
>
> which must be modified to define f=[0 0 0; 0 0 0; 0 0 0]

So which result of these is better?

f(X1) = [10 20;30 40]

f(X2) = [0 0; 0 100]

f(X1) contains a smaller maximum value than f(X2), but more components of
f(X2) are zero than components of f(X1).

You may want to look at FGOALATTAIN, or if you know the 'weighting' you want
to give to each component of the result of your matrix-valued function,
write a scalar-valued function that calls your matrix-valued function and
combines the elements in the correct weighted combination to return a scalar
value for GA or FMINCON to minimize.

--
Steve Lord
slord@mathworks.com
comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ

On 29 Ott, 17:55, "Steven Lord" <sl...@mathworks.com> wrote:
> "Annick " <annick...@yahoo.ca> wrote in message
>
> news:hccfj2$4od$1@fred.mathworks.com...
>
>
>
>
>
> > "Luca Zanotti Fragonara" <Luca_Zano...@libero.it> wrote in message
> > <hccdhe$pa...@fred.mathworks.com>...
> >> "Annick " <annick...@yahoo.ca> wrote in message
> >> <hcccum$id...@fred.mathworks.com>...
> >> > Hi,
>
> >> > I have an objective function f(x) that is a matrix, this objective
> >> > function is evaluated at different point in X(x,y) space
>
> >> > hence each Xi as a value f(Xi) associated with it, this f(Xi) is a
> >> > function of some input vector [ui vi wi zi] which is bounded. The value
> >> > of f(Xi) is influenced by all the i i=1...n
> >> > the f(X1)=f(X1,X2,....Xn)
>
> >> > I want f(Xi)=0 or close to zero for i=1...n by varying the [ui vi wi
> >> > zi] vectors
>
> >> > can I used the simulated annealing tool box to perform this
> >> > optimization, if not which matlab optimization tool can perform this
> >> > analysis
>
> >> > Thanks
>
> >> > Annick
>
> >> You can use also ga or fmincon if I've understood correctly your problem.
>
> > I am not sure that I can use fmincon because the minimization gives only
> > one value
>
> > my
> > f as several values ex f=[2 3 1; 4 5 5; 7 1 3] and I want all the cells of
> > f to be zero or close to zero
>
> > the value of f is obtain from a function that calculates the f values as a
> > function of the input vector [u v w z]
>
> > which must be modified to define f=[0 0 0; 0 0 0; 0 0 0]
>
> So which result of these is better?
>
> f(X1) = [10 20;30 40]
>
> f(X2) = [0 0; 0 100]
>
> f(X1) contains a smaller maximum value than f(X2), but more components of
> f(X2) are zero than components of f(X1).
>
> You may want to look at FGOALATTAIN, or if you know the 'weighting' you want
> to give to each component of the result of your matrix-valued function,
> write a scalar-valued function that calls your matrix-valued function and
> combines the elements in the correct weighted combination to return a scalar
> value for GA or FMINCON to minimize.
>
> --
> Steve Lord
> sl...@mathworks.com
> comp.soft-sys.matlab (CSSM) FAQ:http://matlabwiki.mathworks.com/MATLAB_FAQ

I agree with Steve. For instance, you can define

mynewf=sum(sum(f.^2));

and minimize it. Depending on the problem you have, this should give
to you a matrix f with very small values.

Antonio.

"Annick " <annick219@yahoo.ca> wrote in message <hcccum$idj$1@fred.mathworks.com>...
> Hi,
>
> I have an objective function f(x) that is a matrix, this objective function is evaluated at different point in X(x,y) space
>
> hence each Xi as a value f(Xi) associated with it, this f(Xi) is a function of some input vector [ui vi wi zi] which is bounded. The value of f(Xi) is influenced by all the i i=1...n
> the f(X1)=f(X1,X2,....Xn)
>
> I want f(Xi)=0 or close to zero for i=1...n by varying the [ui vi wi zi] vectors
>
> can I used the simulated annealing tool box to perform this optimization, if not which matlab optimization tool can perform this analysis
>
> Thanks
>
> Annick



If you use fmincon and lb as zeros and upper bound like 0.000001
its just a guess from me try it