Introduction
In the previous lesson, we introduced the idea of differentiating measures.
We studied local averages such as:
$$\frac{1}{\lambda(B(x,r))}\int_{B(x,r)}f(y),d\lambda(y)$$
and asked:
What happens as the ball shrinks to a point?
For continuous functions, intuition suggests that the average value over a very small neighborhood should be nearly equal to the value at the center.
The remarkable discovery of Lebesgue was that this remains true for almost every point, even when the function is merely integrable and may be highly irregular.
This result is called the Lebesgue Differentiation Theorem.
It is one of the deepest and most beautiful theorems in classical analysis.
Many analysts regard it as the true measure-theoretic version of the Fundamental Theorem of Calculus.
The Main Question
Suppose:
$$f\in L^1_{loc}(\mathbb R^n)$$
meaning:
$$f$$
is locally integrable.
Consider the average value of:
$$f$$
over a ball:
$$B(x,r)$$
centered at:
$$x$$
with radius:
$$r$$
The average is:
$$A_r(f)(x)=\frac{1}{\lambda(B(x,r))}\int_{B(x,r)}f(y),d\lambda(y)$$
We ask:
Does this average converge to the actual value of the function?
Continuous Functions
Suppose first that:
$$f$$
is continuous.
Then for sufficiently small:
$$r$$
all points inside:
$$B(x,r)$$
have values close to:
$$f(x)$$
Therefore:
$$A_r(f)(x)\approx f(x)$$
and we obtain:
$$\lim_{r\to0}A_r(f)(x)=f(x)$$
This is unsurprising.
The amazing part is that continuity is not required.
Statement of the Theorem
Let:
$$f\in L^1_{loc}(\mathbb R^n)$$
Then:
$$\lim_{r\to0}\frac{1}{\lambda(B(x,r))}\int_{B(x,r)}f(y),d\lambda(y)=f(x)$$
for Lebesgue almost every:
$$x\in\mathbb R^n$$
What the Theorem Says
Even if:
$$f$$
has infinitely many discontinuities,
even if:
$$f$$
looks extremely irregular,
the local averages still recover the function value at almost every point.
The exceptional set has measure zero.
Why This Is Amazing
Integration appears to destroy local information.
When we integrate, we average values over regions.
One might expect that individual point values become lost.
The theorem says the opposite.
Local averages retain enough information to reconstruct the original function almost everywhere.
Example 1
Consider:
$$f(x)=x^2$$
For small:
$$r$$
the average:
$$\frac{1}{2r}\int_{x-r}^{x+r}t^2,dt$$
approaches:
$$x^2$$
Therefore:
$$\lim_{r\to0}\frac{1}{2r}\int_{x-r}^{x+r}t^2,dt=x^2$$
The theorem agrees with ordinary calculus.
Example 2
Consider:
$$f(x)=|x|$$
Although:
$$f$$
is not differentiable at:
$$0$$
it is continuous.
The theorem gives:
$$\lim_{r\to0}\frac{1}{2r}\int_{-r}^{r}|t|,dt=0$$
which equals:
$$f(0)$$
Example 3
Consider:
$$f=\mathbf1_{[0,1]}$$
At points strictly inside:
$$[0,1]$$
the average tends to:
$$1$$
At points outside:
$$[0,1]$$
the average tends to:
$$0$$
At boundary points:
$$0$$
and:
$$1$$
the behavior is more subtle.
The theorem only guarantees convergence almost everywhere, not everywhere.
Why “Almost Everywhere” Is Necessary
Some functions possess exceptional points where recovery fails.
These exceptional points form a set of measure zero.
Measure theory accepts this because:
sets of measure zero are negligible from the integration perspective.
Thus:
$$\text{a.e. convergence}$$
is the strongest reasonable statement.
Connection to the Fundamental Theorem of Calculus
Recall:
$$F(x)=\int_a^x f(t),dt$$
Classical calculus gives:
$$F’(x)=f(x)$$
for continuous functions.
The Lebesgue Differentiation Theorem extends this dramatically.
Define:
$$F(x)=\int_0^x f(t),dt$$
for:
$$f\in L^1$$
Then:
$$F’(x)=f(x)$$
for almost every:
$$x$$
This is the measure-theoretic Fundamental Theorem of Calculus.
Recovering Densities
Suppose:
$$\nu(A)=\int_Af,d\lambda$$
Then:
$$\frac{\nu(B(x,r))}{\lambda(B(x,r))}=\frac{1}{\lambda(B(x,r))}\int_{B(x,r)}f,d\lambda$$
Applying the theorem yields:
$$\lim_{r\to0}\frac{\nu(B(x,r))}{\lambda(B(x,r))}=f(x)$$
almost everywhere.
Thus the density is recovered from the measure.
Why Radon–Nikodym Really Is a Derivative
Recall:
$$f=\frac{d\nu}{d\lambda}$$
The differentiation theorem shows:
$$\frac{d\nu}{d\lambda}(x)=\lim_{r\to0}\frac{\nu(B(x,r))}{\lambda(B(x,r))}$$
almost everywhere.
This formula looks exactly like an ordinary derivative.
The notation is therefore completely justified.
Maximal Functions
The proof of the theorem relies on one of the most important tools in harmonic analysis.
Given:
$$f$$
define the Hardy–Littlewood maximal function:
$$Mf(x)=\sup_{r>0}\frac{1}{\lambda(B(x,r))}\int_{B(x,r)}|f(y)|,d\lambda(y)$$
This function measures the largest local average around:
$$x$$
The maximal function controls the exceptional sets where convergence might fail.
Why the Proof Is Difficult
The theorem is not proved directly.
Instead:
- Prove it for continuous functions.
- Approximate integrable functions by continuous functions.
- Use maximal inequalities.
- Show exceptional sets have arbitrarily small measure.
The proof is one of the masterpieces of classical analysis.
The Hardy–Littlewood Maximal Theorem
A key ingredient states:
For every:
$$\alpha>0$$
there exists a constant:
$$C_n$$
such that:
$$\lambda({x:Mf(x)>\alpha})\le\frac{C_n}{\alpha}\int |f|,d\lambda$$
This inequality allows analysts to control bad sets quantitatively.
It is one of the foundational results of harmonic analysis.
Geometric Interpretation
Imagine measuring temperature across a country.
Instead of recording temperature at exactly:
$$x$$
you measure average temperature inside increasingly small neighborhoods.
The theorem states:
As the neighborhoods shrink, the average temperature converges to the true temperature at almost every location.
Connection to Probability
Suppose:
$$p(x)$$
is a probability density.
Then:
$$P(B(x,r))\approx p(x)\lambda(B(x,r))$$
for very small:
$$r$$
Thus:
$$p(x)=\lim_{r\to0}\frac{P(B(x,r))}{\lambda(B(x,r))}$$
almost everywhere.
This gives a local interpretation of probability densities.
Connection to Statistics
Kernel density estimation attempts to estimate:
$$p(x)$$
using local averages of observed data.
The theoretical justification ultimately traces back to ideas closely related to the Lebesgue Differentiation Theorem.
Connection to Geometric Measure Theory
The theorem suggests that measures possess local densities.
This idea becomes central in:
- Hausdorff measure
- Fractal geometry
- Rectifiability theory
- Minimal surfaces
Many advanced geometric theories begin here.
Connection to Alain Connes
One of the fundamental ideas of modern analysis is:
Global structures can be understood through local averages.
The Lebesgue Differentiation Theorem embodies this principle.
Later, in noncommutative geometry, Connes develops analogues of:
- measures
- integration
- differentiation
for operator algebras.
The classical theory of local density becomes one of the conceptual foundations upon which these noncommutative ideas are built.
Key Concepts Learned
By the end of this lesson you should understand:
- Local averages are defined by:
$$\frac{1}{\lambda(B(x,r))}\int_{B(x,r)}f(y),d\lambda(y)$$
- The Lebesgue Differentiation Theorem states:
$$\lim_{r\to0}\frac{1}{\lambda(B(x,r))}\int_{B(x,r)}f(y),d\lambda(y)=f(x)$$
almost everywhere.
- The theorem holds for locally integrable functions.
- Local averages recover pointwise values almost everywhere.
- The theorem justifies the Radon–Nikodym derivative as a genuine derivative.
- Maximal functions play a central role in the proof.
- The theorem is a measure-theoretic extension of the Fundamental Theorem of Calculus.
Looking Ahead
Measure Theory Lesson 30: Radon Measures
So far, our measures have largely lived on subsets of Euclidean space. In the next lesson, we generalize the notion of measure to much broader topological spaces and introduce Radon measures, the natural class of measures used in modern analysis, probability theory, differential geometry, and ultimately operator algebras. Radon measures will form one of the first bridges from classical measure theory toward functional analysis and the mathematics that Connes builds upon.
Leave a Reply