Power System Resilience

At work, a colleague was trying to “quantify the resilience of a power system”, to put one number on “resilience”. He even was aware that resilience has multiple aspects: resilient to what — a cyberattack, a hurricane? Is a partial blackout acceptable if everything is back within an hour?

These are different dimensions, not one number. What would it mean to “add” two blackouts?

The most important thing about numbers that I didn't learn at university is:

Not everything is a number!

Once we have found a way to project resilience onto one number, people start improving the number and forget what it was supposed to stand for (Goodhart’s law). In this case, it can even be dangerous to reduce something multidimensional to a scalar.

What a number requires

To behave like a real number, a quantity has to support, among other things, an ordering relation and an operation for combining quantities:

Ordering. For any two aa and bb, exactly one of a>ba > b, a=ba = b, or a<ba < b holds. This relation must also be transitive: if a>ba > b and b>cb > c, then a>ca > c. This gives an order.

Concatenation. We can combine aa and bb. It doesn't matter in which order we combine them (a+b=b+aa + b = b + a), and, however small positive aa is and however large positive bb is, enough copies of aa eventually exceed bb (the Archimedean property).

This goes back to a theorem by Hölder (1901); a modern treatment appears in Foundations of Measurement by Krantz, Luce, Suppes, and Tversky. Only when a structure supports all the required axioms can we represent it faithfully with positive real numbers, preserving both order and concatenation.

Which bike tire should I choose?

The internet's authority on this topic—where I also spend too much of my spare time—is bicyclerollingresistance.com. It gives at least five numbers for each tire: speed, grip, weight, durability, and puncture resistance. They are in direct opposition to one another—more grip costs durability; low rolling resistance costs puncture protection.

Still, there is some order: some tires are worse on every axis. What remains after eliminating those is the Pareto front: the tires for which you can't improve one property without giving up another. (This is similar to the idea in the constraints post: we can't compute the optimum, but we can eliminate bad options.)

What is clear with bike tires is this: on the Pareto front, no tire is better than another until you fix the goal.

In a race, speed is most important. But for the usual coffee ride, one might consider trading a little speed for grip or puncture resistance.

In this example, we can at least measure and quantify the different aspects of a tire. We can keep these measurements as a vector or project them onto a goal to arrive at one number.

Which city should we live in?

We fell into the “number” trap ourselves. Choosing a city to live in, we wrote down categories—job, nature, family and friends—gave each a score, and added them up. The results didn't really make sense.

Of course not: we had added things that cannot be added and tried to rank options that do not fit on a single scale.

You can see this with a thought experiment: suppose the job and nature/family points balance out, making two cities look equal. Now the job offer in one city improves slightly—for example, it pays a little more.

Is that city now the better choice? No! But if the cities were genuinely equal, any improvement should tip the balance.

The philosopher Ruth Chang calls this parity: the options are not better, worse, or equal—they are on a par. Parity is a relation, but it does not rank the options on a single scale.

This is where the choice is ours to make.

What should you do?

Before turning something into one number, ask:

Can I rank it? Can I add it?

If not, remember: Some things cannot be quantified.