# When does the MLE of the Logistic Distribution exist and when is it unique?

I am trying to understand the procedure for finding the MLE of the parameter $$\theta$$ of a Logistic Distribution. The book I am reading has the following derivation:

I am having trouble understanding how the book's analysis shows that MLE exists and is unique (which is discussed in the last paragraph). The book I am reading has not discussed any theorems on either the existence or the uniqueness of MLEs. Is the last paragraph implicitly based on any theorems? In general, how does one know (and verify) that the MLE of a certain parameter of a certain distribution exists and if is exists, that the MLE is unique? Any help is greatly appreciated!

• The second derivative of the log likelihood is strictly negative, hence the log likelihood is concave and therefore has a unique maximum. Apr 18 '20 at 17:02
• You can see this by applying his reasoning to 6 1 9 instead. Take the derivative of that. Now you have 1 minus the term in the sum. But this term is bounded above one, hence the second derivative is negative. Apr 18 '20 at 17:04
• This is a standard analysis based on differential Calculus -- it's not special to MLE. A good intro Calc text will cover every concept and technique used in this quotation.
– whuber
Mar 19 at 23:07

As the Equation 6.1.10 can be likely solved by an iterative numerical method (like Newton-Raphson) using the first derivative and successive point approximations. See this source for operational and convergence details.

The operational question is there a point in the realm of real numbers where the denominator of the first derivative goes to zero. The answer is no.

To demonstrate, if we did set the denominator equal to zero, take the square-root of both sides, and look for solutions indicating a possible singularity:

$$1 + exp( -(x_i - \theta )) = 0$$

Or:

$$exp( -(x_i - \theta )) = -1$$

Or:

$$-(x_i - \theta ) = log(-1)$$

To investigate a solution, I start with some background courtesy of Wikipedia, to quote:

For example, ln(i) = πi/2 or 5πi/2 or -3πi/2, etc.;

So, upon multiplying both sides of the first representation by two, I can successively develop an equivalence:

$$2 \log(i) = log(i^2) = log(-1) = πi$$

where my log function is more precisely defined as the natural log extended to the realm of complex numbers. Side note, I could have also used in place of πi/2 the value -πi/2, in which case the answer is also -πi.

Now, going back to our denominator set equal to zero, a singularity could theoretically arise when $$\theta\ - x_i$$ equals πi. However, this would imply either $$\theta\$$ or $$x_i$$ lie in the complex plane. which they do not.