The AnyLogic Truncated Gamma Distribution is incorrect

As my last post illustrated, AnyLogic's distributions are occasionally incorrect. This keeps you limber as an engineer, because it's always a fun surprise when expensive enterprise software is worse than what free Python libraries provide.

Anyhow, today I am going to illustrate that the AnyLogic Gamma Distribution doesn't do what the documentation says that it should.

There are a couple different call signatures for the Gamma distribution. You can read about them here and here, but essentially you can call it like:

• gamma(alpha, beta, shift)
• gamma(min, max, alpha, shift, stretch)

Alpha and Beta are the parameters, shift and stretch move and scale the distribution, and min and max are truncation bounds.

The documentation assures us that the 5 paramater signature is equivalent to calling the 3 parameter version like so: gamma(alpha, 1, 0) and then multiplying it by the stretch, and moving it by the shift. This means that:

• gamma(0.5, 1, 0)
• gamma(0, 10, 0.5, 0, 1)

should be equivalent. Take a moment to convince yourself of that.

Now let's use the choose probability distribution wizard to visualise both distributions:

As you can see, they're obviously different. Yikes.

The mean of a gamma distribution with alpha and beta parameters is defined as alpha/beta, so the mean for this distribution should be 0.5. As you can see, the 3 parameter invocation is correct, while the 5 parameter version is incorrect. Perhaps it's just wrong. Perhaps the documentation is out of date. You decide.

AnyLogic is the worst.