Error: "The PDF function returned NaN or infinite values" or "The PDF function returned negative or zero values."

Hello all,

I'm trying to make a mixed probability function within the interval [0,1]. All my data is in the interval [0,1]. For my data I thought to do it with 2 possible manners: 1) with a LogNormal and a Normal distributions 2) Using 2 Beta distributions and a uniform one.

In the first case I got as error: "The PDF function returned negative or zero values."

Instead in the second case: "The PDF function returned NaN or infinite values."

I try go through the mle's codes, but it seems impossible for me to understand the mistake.

Below is the code of the second solution (2 Beta distributions and a uniform):

function f=BIMpdf(data,x)

pdf_mix= @(x,p,q,a1,a2,b1,b2) ...
          p*betapdf(x,a1,b1) + (1-p-q)*betapdf(x,a1,b2) + q;

qStart= 0.001;
pStart = .7;

data1= data(data<=0.3);
data2= data(data>0.8);

m= [mean(data1) mean(data2)];
s = [var(data1) var(data2)];

for k=1:2
aStart(1,k)= m(k)*(1-m(k)-s(k))/s(k);

bStart(1,k) =(1-m(k)-s(k))*(1-m(k))/s(k);

end

start = [pStart qStart aStart bStart]

%     p  q a1  a2 b1  b2 

lb = [0  0  0  0  0   0];
ub = [1 1 Inf Inf Inf Inf];

options = statset('MaxIter',600, 'MaxFunEvals',1000);
[paramEsts, pci] = mle(data', 'pdf',pdf_mix, 'start',start, 'lower',lb, 'upper',ub);

f= pdf_mix(x,paramEsts(1),paramEsts(2),paramEsts(3),paramEsts(4),paramEsts(5),parEsts(6));

end

Thanks a lot for your help guys!