I don't see anything wrong with maxima's result. It's not as simple as it could be, but it doesn't look wrong. Using ratsimp makes the expression simpler.
Can you point out what the problem is?
irf(a1,a2,b,c,th1,th2) := c+(1-c)/(1+exp(-(a1th1+a2th2+b)));
diff(log(1-irf(a1,a2,b,c,th1,th2)),a1);
obtains:
-((1-c)th1%e^(-a2th2-a1th1-b))/((%e^(-a2th2-a1th1-b)+1)^2(-(1-c)/(%e^(-a2th2-a1*th1-b)+1)-c+1))
Compare output to Wolfram Alpha. For Wolfram Alpha, I had to rename some of the variables:
d/dx log(1- (c+(1-c)/(1+exp(-(xf+yg+z*h+b)))))
Obtains:
f/(e^(b+fx+gy+h*z)+1)-f
I don't see anything wrong with maxima's result. It's not as simple as it could be, but it doesn't look wrong. Using ratsimp makes the expression simpler.
Can you point out what the problem is?
Sorry, I guess both results are the same.
No problem. Closing this bug.