The forward difference operator (discrete derivative) $\Delta$ is defined as $\Delta f(x) = f(x+1) - f(x)$.

The "discrete $e^x$" / eigenfunction of $\Delta$ is $2^x$.

Since $2^{x+1} - 2^x = (2-1)2^x = 2^x$. The analogue for $e^{\lambda x}$ is $(1+\lambda)^x$ since $$\Delta (1+\lambda)^x = (1+\lambda)^{x+1} - (1+\lambda)^x = (1+\lambda -1)(1+\lambda)^x = \lambda(1+\lambda)^x$$

A Newton series representation for $2^x$ is $\sum\limits_{n=0}^{\infty} \frac{x^{\underline{n}}}{n!}$, using $x^{\underline{n}}$ = $n! \displaystyle {x \choose {n}}$ to denote the falling power. This holds because \begin{equation*}\sum_{n=0}^{\infty} \frac{x^{\underline{n}}}{n!} =\sum_{n=0}^{\infty} {x\choose{k}} = 2^x\end{equation*}

And using the nice analogue of the power rule that $\Delta x^{\underline{n}} = nx^{\underline{n-1}}$ we have

\begin{equation*}\Delta 2^x = \Delta \left( \sum_{n=0}^{\infty} \frac{x^{\underline{n}}}{n!} \right) = \sum_{n=0}^{\infty} \frac{\Delta x^{\underline{n}}}{n!} = \sum_{n=0}^{\infty} \frac{x^{\underline{n}}}{n!} = 2^x\end{equation*}

If I define a "discrete cosine" as \begin{equation*}c(x) = \frac{(1+i)^x + (1-i)^x}{2} = \frac{1}{2}\sum_{n=0}^{\infty}{x \choose {n}}i^n + \frac{1}{2}\sum_{n=0}^{\infty}{x \choose {n}}(-i)^n\end{equation*}

I want to show that this is equal to $\displaystyle \sum_{n=0}^{\infty} (-1)^n \frac{x^{\underline{2n}}}{(2n)!}$ and likewise show

$s(x) := \displaystyle \frac{(1+i)^x - (1-i)^x}{2i} = \sum_{n=0}^{\infty} (-1)^n \frac{x^{\underline{2n+1}}}{(2n+1)!}$

The proof is complete if can get an expression of the form $\displaystyle \sum_{n=0}^{\infty}{x \choose {2n}}(i)^{2n}$ for $c(x)$ which is similar to the above expression which I get by a binomial expansion, and the same for $s(x)$

I also recall to have seen other functions $s^*(x), c^*(x)$ with the same property of solving $\Delta^2 f = -f$ in a pdf I can no longer find, but I remember them looking extremely different, involving a $2^x$ term and the respective "normal" trig functions. The closed form I think was roughly in the form $2^{kx}$ $\cos (\alpha x + \beta)$ for $c^*(x)$ and $2^{kx}$ $\sin(\alpha x + \beta)$ for $s^*(x)$ (same constants if I remember correctly).

I would like to know how this was obtained (and what the formula actually is), though this is more of a side question I would be grateful if anyone else knows of this and its proof.

  • 1
    $\begingroup$ For the first question, do the same as for the regular sine, cosine and derivative: Take $\Delta$ on both sides repeatedly and evaluate at $x=0$. $\endgroup$ – user517969 Jan 2 '18 at 21:57
  • $\begingroup$ @plumbus you are indeed correct, although I am still curious about other proofs. $\endgroup$ – Theo Diamantakis Jan 3 '18 at 1:58
  • 2
    $\begingroup$ As a sanity check... you do realize that in your "sum of two summations" formula for cosine, the odd indexed terms cancel out and the even index terms are the same? $\endgroup$ – user14972 Jan 3 '18 at 2:50
  • $\begingroup$ Ah I see what you mean, the proof follows from that simple observation, and you can do the same trick with sine. If you post a few words as an answer I'll give you the reputation and accept it haha. Thanks. $\endgroup$ – Theo Diamantakis Jan 3 '18 at 2:52

Given a power series, there is a standard trick to isolate its even or its odd indexed terms. If

$$ f(x) = \sum_{n=0}^{\infty} a_n x^n $$


$$ f(-x) = \sum_{n=0}^{\infty} a_n (-x)^n = \sum_{n=0}^{\infty} (-1)^n a_n x^n $$

so that both series have the same even terms and negated opposite terms. Thus

$$ \frac{f(x) + f(-x)}{2} = \sum_{\substack{n=0 \\ n \text{ even}}}^\infty a_n x^n = \sum_{m=0}^{\infty} a_{2m} x^{2m}$$

$$ \frac{f(x) - f(-x)}{2} = \sum_{\substack{n=0 \\ n \text{ odd}}}^\infty a_n x^n = \sum_{m=0}^{\infty} a_{2m+1} x^{2m+1}$$

The combination of sums you're asking about happens to be precisely of this form:

$$ \frac{1}{2}\left( \sum_{n=0}^{\infty}{x \choose {n}}i^n + \sum_{n=0}^{\infty}{x \choose {n}}(-i)^n\right) = \sum_{m=0}^{\infty}{x \choose {2m}} i^{2m} $$


After lots of fruitless searching I managed to guess the mysterious $c^*(x), s^*(x)$ via trial and error.

The functions $$2^{\frac{x}{2}} \sin(\frac{\pi}{4}x + 2) $$ and $$2^{\frac{x}{2}} \cos(\frac{\pi}{4}x + 2) $$

Satisfy $\Delta s^*(x) = c^*(x)$ and $\Delta c^*(x) = -s^*(x)$, but $s^*(x) \neq s(x)$ and $c^*(x) \neq c(x)$. Since $\Delta f = \Delta (f+g)$ when $g$ is a function which has period $1$ i.e $g(x+1) = g(x)$ (because $\Delta g = 0$ and $\Delta$ is linear) $s(x), s^*(x)$ must differ by a $1$ periodic function.

$s(x)$ and $c(x)$ can be still considered the "true" discrete trigonometric functions because of the power series analogy, but this is just a matter of taste.

For another interesting choice see this question and answer, where the periodic property is kept but not the similar looking power series https://math.stackexchange.com/a/1439726/462531


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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