Helpful limitations

Through my work on power-system optimization, I've been thinking more about constraints. The first thought might be: constraints are restrictions; constraints are bad. But interestingly, this isn't always true.

Binding constraints

Let's dive into constraints. Take this as an example:

min⁡f(x)=xs.t.x≥2,  x≥3.\min f(x)=x \quad \text{s.t.} \quad x \ge 2,\; x \ge 3.

Here, x≥3x \ge 3 is the important one — it defines the optimum and pushes it away from where we'd like it to be. Without it, xx could go lower. The constraint x≥2x \ge 2 is present, but it does nothing. If we can remove x≥3x \ge 3, the optimum moves to 2 and the solution gets better.

It's interesting to look for such constraints in the real world. If we try to optimize for speed — for example, getting somewhere faster or finishing something faster — the binding constraint might be the one traffic light on the route or the one process at work that always takes days. No matter how much we improve everything else, the whole process is limited by this. Naturally, this is what we should work on (if it lies within our reach).

While the traffic light is obvious, the same is true of things that are harder to notice. Happiness, a partnership, a career — here, too, there is often one thing holding everything back.

Strengths or weaknesses

This should answer the question: Should I work on my strengths or on my weaknesses? It's worth actively searching for the binding constraint and working on that one. (This is known as bottleneck thinking; see Goldratt's Theory of Constraints.)

It's important to realize that weakness here means "the thing holding you back" from achieving what you want, not a weakness in the sense of "I cannot juggle."

Constraints as information

So far, the story is simple: find the constraint that holds you back and remove it.

But that still leaves a puzzle I didn't understand for a long time. A lot of good work seems to happen not with fewer constraints, but with more.

Many photographers shoot with a single fixed focal length. The old iPod was built around a scroll wheel and almost no buttons. Some people work best with a deadline breathing down their necks. Spielberg showed little of the shark in Jaws because the mechanical shark kept breaking, but showing less of the shark made the film even scarier.

Why does this work? Why can such strong limitations produce great work? Why isn't the result simply better with full freedom? Zoom lenses, buttons everywhere, sharks everywhere?

Cutting-plane method and branch-and-bound

In mathematical optimization, there is the idea of the cutting-plane method. Say we're looking for an integer-valued solution to an optimization problem: x∈Zx \in \mathbb{Z}. That's really hard, so first we solve an easier version in which fractional values of xx are allowed. The answer might come back as 5.5. So we add a constraint — a cut — x<5x < 5 or x>6x > 6, chosen to remove that fractional answer and, ideally, a whole region of similarly invalid answers. Then we solve the relaxed problem again.

Step by step, we carve the space down until what's left is an integer solution.

The added constraint isn't a restriction. It's information: "not there". If we already know the answer is small, we can safely add x<5x < 5 — and never have to search near x≈100x \approx 100 again.

This sheds new light on the earlier examples. It's still entirely possible to create a photographic masterpiece with a 50mm lens and to write a great article in an evening. Spielberg didn't need to show the shark, and the iPod didn't need more buttons to be better.

In art, design, and life, we usually move through spaces with infinitely many options. Full freedom sounds like a gift, but really, it's paralysis. Every good option is available — but so is every bad one.

So which is it — do constraints hold us back or push us forward? The binding constraint stands between us and a goal we can already see. The chosen constraint works when we can't see the goal yet.

This makes it clear how we can actively use constraints to help us. A restriction is often much easier to find than the optimum. So the question isn't this: How much freedom do I need to reach the goal? It's this: Under which (strong!) restriction is the goal still reachable?