5
$\begingroup$

I was wondering whether there is a Euler–Maclaurin formula of sorts for expressions such as $$ \sum_{x \in [a,b]^d\cap \mathbb{Z}^d} f(x) - \int_{[a,b]^d}f(x) $$ where $d\ge 2$ is an integer, $a,b \in \mathbb{R}$ and $f:\mathbb{R}^d \longrightarrow \mathbb{R}$ is a smooth function in $[a,b]^d$. I am particularly interested in such expansion with a control of the error term.

I appreciate any reference or suggestions.

$\endgroup$

2 Answers 2

6
$\begingroup$

See e.g. Y. Karshon, S. Sternberg, and J.Weitsman. The Euler-Maclaurin formula for simple integral polytopes and Y. Karshon, S. Sternberg, and J. Weitsman. Euler-Maclaurin with remainder for a simple integral polytope.

For an alternative summation formula -- in terms of integrals only, without using derivatives -- see Approximating sums by integrals only: multiple sums and sums over lattice polytopes.

In these papers, you will also find explicit expressions for and/or simple bounds on the remainder.

$\endgroup$
4
$\begingroup$

sorry for necromancing this thread but I wanted to give an elementary (with physics style abuse of notation) answer to this:

Yes there are some simple generalizations that support this. We need to recall where the Euler-Maclaurin formula comes from:

Let $I$ denote the identity operator. Let $H$ denote the shift operator $H[f] = f(x+1)$. Then (with abuse of notation) we can compute the following:

$$ \frac{I}{I-H} = I + H + H^2 + H^3 ... = f(x)+f(x+1)+f(x+2)+f(x+3)...$$

So then it follows that

$$ - \frac{I}{I-H} |_{x=a}^{x=b} = f(a)+f(a+1)+ \ ... \ + f(b-1) $$

Now if you recall that $H = e^{\frac{d}{dx}} $

We then have

$$ \sum_{k=a}^{b-1} f(x) = - \frac{I}{I-e^{\frac{d}{dx}}} |_{x=a}^{x=b} $$

Now recall on the right hand side that

$$ - \frac{1}{1-e^x} = \frac{1}{x} - \frac{1}{2} + \frac{1}{12}x + ... $$

So we conclude then that:

$$ - \frac{I}{I-e^{\frac{d}{dx}}} |_{x=a}^{x=b} = \int_{a}^{b} f dx - \frac{1}{2}(f(b)-f(a)) + \frac{1}{12} (f'(b)-f(a)) + ... $$

Which is the Euler Maclaurin formula.

Now we go to the multivariable case. Suppose we have a function $f(x,y)$ we have shift operators $H_x = H(x+1,y)$ and $H_y = H(x,y+1)$

It's easy to see that $H_x, H_y$ commute so that with abuse of notation we can write

$$ \sum_{n=a}^{b-1} \left[ \sum_{k=c}^{d-1} f(n,k) \right] = \frac{I}{I-H_x } \frac{I}{I-H_y}|_{x=a}^{x=b} |_{y=c}^{y=d} = \frac{I}{I-e^{\frac{\partial}{\partial x}}} \frac{I}{I-e^{\frac{\partial}{\partial y}}}|_{x=a}^{x=b} |_{y=c}^{y=d} $$

Thereby giving you that:

$$ \sum_{n=a}^{b-1} \left[ \sum_{k=c}^{d-1} f(n,k) \right] = \int_{c}^{d} \left( \int_{a}^{b} f(x,y) \partial x - \frac{f(b,y)-f(a,y)}{2} + \frac{ \partial_x[f]_{x=b} - \partial_x[f]_{x=a} }{12} + ... \right) \partial y + \\ -\frac{1}{2} \left( \int_{a}^{b} f(x,y) \partial x - \frac{f(b,y)-f(a,y)}{2} + \frac{ \partial_x[f]_{x=b} - \partial_x[f]_{x=a} }{12} + ... \right)|_{y=c}^{y=d} + ... $$

Which is just the Euler Maclaurin Formula for variables $x$ and $y$ composed with each other.

In general moving to $\mathbb{Z}^d$ follows naturally from this by just consider each variable $x_1 ... x_d$ separately and composing them just like we did for $\mathbb{Z}^2$ above.

As a more general corollary. If we want to evaluate

$$ f(x+c_0) + f(x+c_1) + f(x+c_2) + ... $$

Then we can let $T(x) = \sum_{n=0}^{\infty} x^{c_n}$. Then the operator $T(e^{\frac{d}{x}})$ when expanded via the Laurent series for $T(e^x)$ will give you the corresponding euler maclaurin formula.

As an example if we want to evaluate

$$ f(x) + f(x+1) + f(x+\sqrt{2}) + f(x+\sqrt{3}) + ... $$

We look at $\sum_{n=0}^{\infty} x^{\sqrt{n}} = \frac{\Gamma(1+2)}{(-z)^2} + \sum_{k=0}^\infty \zeta\left(-\frac{k}{2}\right)\frac{z^k}{k!}$ as per here: and therefore we conclude

$$ f(x+\sqrt{a}) + f(x+\sqrt{a+1}) + f(x+\sqrt{a+2}) + ... f(x+\sqrt{b-1}) = \\ \underbrace{-2\int \int f(x) dx}_{\text{bounds selected carefully}} - \frac{f(b)-f(a)}{0!2} - \frac{\zeta(-\frac{1}{2})(f'(b)-f'(a))}{1!} + \frac{(f''(b)-f''(a))}{2!12} + ... $$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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