always this message : The first-order optimality measure, 0.0, is less than options

hello all,

Recently I came to the forum to ask for help about the functions of the tool box "optimization". I used "lsqnonlin" "lsqcurvefit" "finminuc" and every time I the same result:

my optimization stops after one iteration and the result: The first-order optimality measure, 0.000000e +000, IS = Less than options.TolFun 1.000000e-006.

after reading the documentation, it understand that is possible that i have a local minimum, yet when I change a little the starting values i have always the same problem: it seems odd to me that every time I put a game parameters of the algorithm says I have a local minimum?!?

can you tell me your opinion on that and how can I solve this. thank you very much

[Jan]

Walter asked you already to show the relevant parts of the code. Without seeing them, only wild guessing is possible.

Answer 1

[21did21]

Gamma=0.21;
A=5100;
B=-0.07;
parametresInitiaux=[Gamma A B];
lb=[0.13 3500 -1.5];
ub=[2.6 8000 5];
options = optimset('Display','iter','MaxIter',40,'TolX',1e-16,'TolFun',1e-25)
[x,resnorm] = lsqnonlin(@calculs,parametresInitiaux,lb,ub,options);

and for calculs:

function RESIDU=calculs(parametresI)
[Rm_num,Fv_num] = WriteCalculTreatment(parametresI(1),parametresI(2),parametresI(3));
RmMine=Rm_num;
FvMine=Fv_num;
residuVect=abs(Rm_exp-Rm_num)+abs(Fv_exp-Fv_num);
RESIDU=residuVect';

the function "WriteCalculTreatment" launch calculations from a software to get Rm_num et Fv_num

when i use fminuc i do this in my function "calcul"

residuVect=(Rm_exp-Rm_num)'*(Rm_exp-Rm_num)+(Fv_exp-Fv_num)'*(Fv_exp-Fv_num);
RESIDU=residuVect;

[Jan]

Please post the code by editing the original image and use code formatting as explained in the "Markup help" link. In the current form, the code is not readable.

[Walter Roberson]

We would need to see WriteCalculTreatment to determine more.

What also need to see the functions Rm_exp and Fv_exp. We deduce they are functions because if they were variables they would have to be initialized within the scope of calculs.

Why do you bother to set RmMine or FvMine ? They are not global or persistent variables, so their value is going to be thrown away (unless they happen to be shared within a scope that you do not show us.)

Answer 2

[21did21]

thank for your help !

Jan : i have already put the matlab code with code formatting...?

Walter : Rm_exp and Fv_exp are global variables. RmMine and FvMine are not necessary.

the function "writeCalculTreatment" is this :

function [Rm_num,Rv_num]=WriteCalculTreatment(Gamma1,A1,B1,DO_mg,DO_si,timeT6)

currentFolder=cd('E:\methodeINVERSE'); T6folder = cd('E:\methodeINVERSE\T6'); T200folder = cd('E:\methodeINVERSE\T200'); T300folder = cd('E:\methodeINVERSE\T300'); T400folder = cd('E:\methodeINVERSE\T400'); T500folder = cd('E:\methodeINVERSE\T500'); T560folder = cd('E:\methodeINVERSE\T560');

currentFolder='E:\methodeINVERSE';
T6folder = 'E:\methodeINVERSE\T6';
T200folder = 'E:\methodeINVERSE\T200';
T300folder = 'E:\methodeINVERSE\T300';
T400folder = 'E:\methodeINVERSE\T400';
T500folder = 'E:\methodeINVERSE\T500';
T560folder = 'E:\methodeINVERSE\T560';
writeT6DataFile(Gamma1,A1,B1,DO_mg,DO_si,timeT6,T6folder,currentFolder);
writeDataFile(Gamma1,A1,B1,DO_mg,DO_si,T200folder,currentFolder);
writeDataFile(Gamma1,A1,B1,DO_mg,DO_si,T300folder,currentFolder);
writeDataFile(Gamma1,A1,B1,DO_mg,DO_si,T400folder,currentFolder);
writeDataFile(Gamma1,A1,B1,DO_mg,DO_si,T500folder,currentFolder);
writeDataFile(Gamma1,A1,B1,DO_mg,DO_si,T560folder,currentFolder);
%%on lance calcul T6 et on transfert ces resultats dans autres dossier
launchT6calcul(T6folder,currentFolder);
%%on ecrit toutes les distributions initiales et lance les calculs
%associés
writeInputFile(T200folder,currentFolder);
writeInputFile(T300folder,currentFolder);
writeInputFile(T400folder,currentFolder);
writeInputFile(T500folder,currentFolder);
writeInputFile(T560folder,currentFolder);
%%detection de la fin des resultats
finishCalculation(T200folder,currentFolder);
finishCalculation(T300folder,currentFolder);
finishCalculation(T400folder,currentFolder);
finishCalculation(T500folder,currentFolder);
finishCalculation(T560folder,currentFolder);
%%traitement des resultats
resultRayon=zeros(6,1);
resultFraction=zeros(6,1);
[R1,F1]=resultTreatment(T6folder,currentFolder);
resultRayon(1,1)=R1;
resultFraction(1,1)=F1;
[R2,F2]=resultTreatment(T200folder,currentFolder);
resultRayon(2,1)=R2;
resultFraction(2,1)=F2;
[R3,F3]=resultTreatment(T300folder,currentFolder);
resultRayon(3,1)=R3;
resultFraction(3,1)=F3;
[R4,F4]=resultTreatment(T400folder,currentFolder);
resultRayon(4,1)=R4;
resultFraction(4,1)=F4;
[R5,F5]=resultTreatment(T500folder,currentFolder);
resultRayon(5,1)=R5;
resultFraction(5,1)=F5;
[R6,F6]=resultTreatment(T560folder,currentFolder);
resultRayon(6,1)=R6;
resultFraction(6,1)=F6;
%%renvoi des donnees
Rm_num=resultRayon;
Rv_num=resultFraction;

here i write data for the software and i launch calculs with the software and i retrieve results. => when i use my programm without lsqnonlin it works nice (my resultats are the results which i have directly with the software) => when i use lsqnonlin it says that the gradient of my function is null ??? (but if i change the value manually i can see what i haven't a minimum...)

[Walter Roberson]

No variable is global unless it has been declared as global in the scope of the routine that is using it. The most obvious way to do that is with a "global" statement right inside calculs; the other methods require that the various routines be "nested" within the routine that declares the variable as global.

Your WriteCalculTreatment function needs all 6 of the variables declared in its header, as it attempts to pass all 6 of them to WriteT6DataFile() . Your calcul() routine is, however, only passing three parameters to WriteCalculTreatment

Answer 3

[21did21]

1°) thank walter i will see this

2°) i have one general question :

how i can increase the discretisation for the gradient in all this optimization function ?