There is a question that runs underneath a surprising amount of mathematics: when can you flip two operations and get the same answer?
It shows up everywhere. Can I differentiate inside an integral? Can I take the limit of a sum by summing the limits? Can I swap an integral and a derivative? Each of these is a different surface manifestation of the same underlying question, and the answer is almost never "always."
The reason they're all the same question is that limits are hiding everywhere. A derivative is a limit. An integral is a limit. An infinite sum is a limit. So "can I swap these two operations" almost always reduces to "can I swap two limits"...and that is where things get interesting.
Consider a sequence of functions fₙ on the real line. Each one is a spike of height n and width 1/n, sitting just to the right of zero. The area under each spike is always exactly 1. But as n grows, the spike gets taller and narrower, and in the limit, it collapses to a single point which has area zero.
So, the limit of the integrals is 1. But the integral of the limit is 0. They disagree.
This is not a pathological example cooked up to cause trouble. It is the generic behavior when you try to move a limit inside an integral without checking conditions. Mass escaped to a point, and the integral couldn't see it anymore.
Three theorems tell you when swapping a limit and an integral is safe.
The first is the Monotone Convergence Theorem. If your functions are nonnegative and increasing, each fₙ ≤ fₙ₊₁ everywhere then you can swap freely. The functions are building up to their limit without any mass escaping.
The second is Fatou's Lemma. Weaker than MCT and it only gives you a one-sided inequality. The integral of the limit is at most the limit of the integrals. The spike example saturates this: 0 ≤ 1. Mass can still escape, but Fatou tells you which direction the inequality goes.
The third is the Dominated Convergence Theorem, which is the one you actually reach for in practice. If there exists an integrable function g that dominates every fₙ meaning |fₙ| ≤ g everywhere; then you can swap. The dominating function prevents mass from escaping to infinity. The spike fails this condition: no single integrable function can dominate spikes of unbounded height.
A sum is an integral against counting measure. This sounds like a technicality, but it means the three theorems above apply immediately to infinite sums. The question "can I swap the sum and the limit" is the same question, with the same answer: yes, if dominated, yes if nonneg and increasing, be careful otherwise.
Pointwise convergence is not enough to swap a limit and a derivative. You need uniform convergence the functions have to converge at the same rate everywhere, not just at each individual point.
The failure mode is a sequence of functions that converges pointwise to zero, but whose derivatives oscillate wildly and don't converge to zero at all. The limit of the derivatives and the derivative of the limit come apart.
Differentiating under the integral sign is Leibniz's rule, and it is the one case where the conditions are relatively mild and the payoff is enormous. If the integrand and its derivative with respect to the parameter are both continuous and bounded, you can move the derivative inside the integral.
In the language of the partition function: if you write Z(β) = ∫ exp(−βE) dμ and want to differentiate with respect to β to extract the mean energy, you are differentiating under the integral sign. The conditions are almost always satisfied for well-behaved energy functions. This is why physicists do it without thinking and mathematicians occasionally wince.
No comments:
Post a Comment