Linearity

And I'm like, if you think that, you're fired.

Linear, quadratic, algebraic bullshit.

Did you just do,

no.

But

the question is, what algorithm am I using to let it roll through, right? What? Linear

summation.

Linear would give me much smaller.

So I think because the improvement is going to be greater in the early years, I couldn't use linear.

It would take too long to get to positive if you start at minus 36.7. What's my current margin? Minus 36.7.

But then what's important is that the current ID varies linearly with the incident optical power.

So you get a linear detector.

Linear in power, not linear in amplitude of the field, but linear in the optical power.

Then you can further increase the reverse bias, so make it more negative.

And then you get into the nonlinear regime.

thanks to, since we have a linear model for our state transition equation, the generic model is indeed a Gaussian.

That is what we have here.

Linear model corresponds to, implies that the probabilistic generic model is Gaussian.

So this is our inductive step.

This follows from the assumption about a linear transition model.

which is nonlinear, and hence destroys the property that when we push the Gaussian random variable through it, what we get as an output is no longer Gaussian.

So what we do is to view this nonlinear transformation, we approximate this nonlinear transformation with a linear one by using the concept of linearization.

Linearization, as you're already familiar with because calculus is a prerequisite, says that a function f that is differentiable can be approximated as f of x bar plus derivative of f with respect to x evaluated at x bar times x minus x bar, where x bar is the linearization point.