An Approximation Theorem for Functions (old)

Q: What is a “delta sequence” 3k, and why does it matter for approximation?

A delta sequence 3k is a family of functions on 3n with three properties: (1) 3k(y) 0 for all y, (2) 3k integrates to 1 over 3n for every k, and (3) the functions concentrate near the origin—given any epsilon > 0, the integral of 3k over the complement of the epsilon-ball around 0 tends to 0 as k 3. This matters because convolution (f * 3k)(x) becomes a weighted average of f(x-y) with weights that collapse toward y = 0, forcing f * 3k to approach f.

Q: How does the proof split the convolution error integral into two regions?

After rewriting the error as an integral of 3k(y)(f(x) - f(x-y)), the integral is split into ||y|| < delta/2 and its complement. On the small ball, uniform continuity bounds |f(x) - f(x-y)| by , so that part of the integral contributes at most . On the complement, the delta-sequence property ensures the weighted mass of 3k there becomes arbitrarily small as k grows; with the explicit compactly supported 3k, choosing k large enough can even make that complement contribution exactly 0.

Q: What does the supremum norm convergence mean in practical terms?

It means the approximation error is uniformly small across the entire compact set A: ||f - f * 3k||3,A 3 0. So a single k suffices to make the difference between f and its smooth approximation less than any prescribed everywhere on A, not just at individual points.

TL;DR

A normalized C3 bump function with compact support can be scaled to form a delta sequence 3k whose mass concentrates near 0 as k 3.

Briefing Cornell Notes

Briefing

A continuous function on 3n can be uniformly approximated on any compact set by smooth (C3) functions using convolution with a carefully constructed “delta sequence.” The key payoff is that for every compact domain A, the smoothed functions f * 3k converge to f in the supremum norm, meaning the approximation error shrinks uniformly across A—not just point by point. This matters because it turns approximation by smooth functions into a reliable, quantitative tool for analysis on compact regions.

The construction starts with a standard smooth bump function 3a on 3n that is nonnegative, infinitely differentiable, and has compact support. It is built from an exponential expression that decays smoothly to zero outside a unit ball, then normalized by a constant C so that its integral over 3n equals 1. From this “standard mollifier,” a whole family 3k is generated by shrinking the bump’s support as k increases: the peak is scaled up while the input is scaled so the effective radius becomes 1/k. This family forms a delta sequence because it satisfies three defining properties: (1) nonnegativity, (2) total mass 3k integrates to 1 for every k, and (3) concentration near the origin—outside any fixed epsilon-ball around 0, the integral of 3k tends to 0 as k 3.

With the delta sequence in hand, the approximation theorem targets a continuous function f on 3n (the transcript even notes f may take complex values). Fix a compact set A 3 3n and consider the supremum norm of the error on A: ||f - 3k * f||3,A. Convolution with 3k produces a C3 function because 3k is smooth. The theorem then claims that this supremum norm goes to 0 as k 3, giving uniform convergence on A.

The proof uses uniform continuity on compact sets. Since f is continuous on compact A, it is uniformly continuous there: for any > 0, there exists a single (independent of the point in A) such that whenever ||y|| < , the values f(x) and f(x-y) differ by less than . The convolution error at a point x in A is written as an integral of 3k(y) (f(x) - f(x-y)). The normalization property (integral of 3k equals 1) allows f(x) to be inserted and removed cleanly, leaving only differences f(x) - f(x-y) inside the integral.

The integral is then split into two regions: where ||y|| < /2 and where ||y|| /2. On the small ball, uniform continuity forces the integrand difference to be < , so the contribution is controlled. On the complement, the delta-sequence concentration property ensures the mass of 3k there becomes negligible for large k. For the explicit compactly supported 3k used in the construction, the complement contribution can even be made exactly zero by choosing k large enough that the support of 3k lies entirely inside the small ball. Taking the supremum over x 3 A preserves the bound, yielding ||f - 3k * f||3,A 3 0.

In short: smooth mollifiers built from normalized bump functions concentrate near 0 as k grows, and uniform continuity on compact sets converts that concentration into uniform approximation by C3 functions.

Cornell Notes

On any compact set A 3 3n, a continuous function f can be uniformly approximated by smooth functions via convolution with a delta sequence 3k. The delta sequence is a family of nonnegative C3 functions with total mass 1 and whose mass concentrates near the origin: for any epsilon-ball around 0, the integral of 3k outside that ball goes to 0 as k 3. The approximation uses uniform continuity of f on compact A: when ||y|| is small, f(x-y) stays within of f(x) for all x in A. The convolution error becomes an integral of 3k(y)(f(x)-f(x-y)), which is split into a “small y” region controlled by uniform continuity and a “large y” region controlled by the delta-sequence concentration. The result is ||f - 3k * f||3,A 3 0, i.e., uniform convergence on A.

What is a “delta sequence” 3k, and why does it matter for approximation?

A delta sequence 3k is a family of functions on 3n with three properties: (1) 3k(y) 0 for all y, (2) 3k integrates to 1 over 3n for every k, and (3) the functions concentrate near the origin—given any epsilon > 0, the integral of 3k over the complement of the epsilon-ball around 0 tends to 0 as k 3. This matters because convolution (f * 3k)(x) becomes a weighted average of f(x-y) with weights that collapse toward y = 0, forcing f * 3k to approach f.

How is the smooth bump function 3a on 3n constructed and normalized?

The construction starts from a standard smooth bump that decays to 0 outside a unit ball using an exponential term of the form exp(1/(||x||^2 - 1)) (with the exponent arranged so the function is 0 outside the ball). Then it is multiplied by a positive constant C chosen so that the integral over 3n equals 1. This normalization ensures convolution with 3a preserves the “mass” needed for approximation.

Why does convolution with 3k produce a C3 function?

Because 3k is infinitely differentiable (C3) and has compact support, convolving it with a continuous f yields a smooth function. Intuitively, derivatives can be passed through the integral since 3k is smooth and integrable, so the resulting function inherits C3 regularity.

Where does uniform continuity enter the proof, and why is compactness essential?

Uniform continuity is guaranteed because f is continuous on a compact set A. That means for any > 0 there exists a single delta > 0 such that for all x 3 A and all y with ||y|| < delta, one has |f(x) - f(x-y)| < . Without compactness, delta might depend on x, and the uniform supremum-norm estimate would fail.

How does the proof split the convolution error integral into two regions?

After rewriting the error as an integral of 3k(y)(f(x) - f(x-y)), the integral is split into ||y|| < delta/2 and its complement. On the small ball, uniform continuity bounds |f(x) - f(x-y)| by , so that part of the integral contributes at most . On the complement, the delta-sequence property ensures the weighted mass of 3k there becomes arbitrarily small as k grows; with the explicit compactly supported 3k, choosing k large enough can even make that complement contribution exactly 0.

What does the supremum norm convergence mean in practical terms?