Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Borel lemma states that for $x_0 \in \mathbf{R}$ and for a real sequence $(a_n)_{n \in \mathbf{N_0}}$ there exists a smooth function $f: \mathbf{R} \rightarrow \mathbf{R}$ such that $f^{(n)}(x_0)=a_n$ for $n \in \mathbf{N_0}$.

However is it true for $x_1, x_2 \in \mathbf{R}, x_1 \neq x_2$, and for real sequences $(a_n)_{n \in \mathbf{N_0}}, (b_n)_{n \in \mathbf{N_0}}$ there exists a smooth function $f: \mathbf{R} \rightarrow \mathbf{R}$ such that $f^{(n)}(x_1)=a_n$, $f^{(n)}(x_2)=a_n$ for $n \in \mathbf{N_0}$ ?


share|cite|improve this question
By using bump functions and Borel's lemma, we find a function $f_1$ whose derivatives at $x_1$ match $a_n$, and which is identically zero outside of a small neighorhood (small enough that it does NOT contain $x_2$), and similarly a $f_2$ that does what we want at $x_2$ and is $0$ outside of a small neighborhood. Then $f_1+f_2$ does what we want. In general, bump functions and partitions of unity allow you to make smooth functions that have specified local behavior. – Aaron Sep 21 '11 at 20:13
up vote 4 down vote accepted

Yes. Take a bump function $\zeta_1$ centered at $ x_1$ and a bump function $\zeta_2$ centered at $x_2$ such that $\mathrm{supp}(\zeta_1) \cap \mathrm{supp}(\zeta_2)=\varnothing$. Use Borel's lemma to find $f_1, f_2 \in C^\infty(\mathbb{R})$ s.t. $f_1^{(n)}(x_1)=a_n, f_2^{(n)}(x_2)=b_n$. Then $f=\zeta_1f_1+\zeta_2f_2$ has the desired property.

share|cite|improve this answer

Since Giuseppe has already given an answer, I will give the details of the extension.

Let $A\subset \mathbb R$ a set which consist of isolated points and for each $a\in A$, $\{x(a)_n\}$ a sequence of real numbers. We can find a smooth function $f$ such that for all $a\in A$ and $n\in\mathbb N$, we have $f^{(n)}(a)=x(a)_n$.

Since $\mathbb R$ is separable, this set has to be countable, hence we can write $A=\{a_n,n\in\mathbb N\}$. We can write $b_{n,k}=x(a_n)_k$ for each $n$ and $k$. For each $n$, we consider $r_n $such that $\left]a_n-3r_n,a_n+3r_n\right[\cap A=\{r_n\}$. Let $g_n$ be a bump function, $g_n=1$ on $\left]-r_n,r_n\right[$ and $\mathrm{supp}g_n\subset \left[-2r_n,2r_n\right]$. Put $$f_{n,j}(x)= \frac{b_{n,j}}{j!}(x-a_n)^jg_n\left(\frac{x-a_n}{\varepsilon_{n,k}}\right)$$ where $\varepsilon_{n,k}=\frac 1{1+|b_{n,k}|}\frac 1{4^kM_{n,k}}$, with $\displaystyle M_{n,k}:=\max_{0\leq j\leq k}\sup_{x\in\mathbb R}|g_n^{(j)}(x)|$. Note that $f_{n,k}$ has a compact support, and putting for all $n$: $\displaystyle h_n(x):=\sum_{k=0}^{+\infty}f_{n,k}(x)$, $h_n$ is a smooth function. Indeed, for $n$ fixed and $k\geq d+1$ we have $\sup_{x\in\mathbb R}|f_{n,k}^{(d)}|\leq \frac 1{2^k}$, hence the normal convergence of the series on the whole real line for all the derivatives. A (boring but not hard) computation which uses Leibniz rule gives that we have $h_n^{(k)}(a_n)=b_{n,k}$.

Now, put $$f(x):=\sum_{n=0}^{+\infty}\sum_{j=0}^{+\infty}f_{n,j}(x).$$ Since the supports of $h_n$ are pairwise disjoint, the series in $n$ is convergent to a smooth function.

We will note that we cannot extend the result to sets $S$ which contain a point which is not isolated, because the continuity of the derivative gives a restriction on the sequence $\{x(s)_n\}$ for the $s$ in a neighborhood of the non-isolated point. Namely, if $s_0$ is a non-isolated point and $\{s_n\}\subset S$ a sequence which converges to $s_0$ then the sequence $\{x(s_n)_0\}$ has to converge to $x(s_0)_0$ (hence we can't get all the sequences).

share|cite|improve this answer
Maybe I misunderstood something. I have one doubt. Let's assume that $A$ has in $\mathbb R$ an accumulation point, say $x_0$ ($x_0$ is not by assumption in $A$). The sum on RHS of the formula $f(x_0)=\sum_{n=0}^\infty h_n(x_0)$ is finite but in arbitrary neighbourhood $U$ of $x_0$ there is infinite many nonzero terms. For differentiability of $f$ in $x_0$ we need uniformly convergence on $U$ of all series $\sum_{n=0}^\infty f_n^(k)(x)$ ($k=0,1,2,...$). – Alex Oct 28 '13 at 17:01

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.