Averaging for nonlinear systems

87 Views Asked by At

I am trying to figure out how the following result has been obtained.

Consider a function $J:\mathbb{R} \longrightarrow \mathbb{R}$ and a dynamical system:

$$ \dot{ \hat{x} }(t) = k a \sin ( \omega t ) J( \hat{x}(t) + a \sin ( \omega t) ) $$

with constant parameters $k, a, \omega$.

Consider Taylor series expansion:

$$ J( \hat{x}(t) + a \sin ( \omega t) ) = J( \hat{x}(t) ) + \frac{ \partial J }{ \partial \hat{x} } \Bigg|_{ \hat{x} = \hat{x}(t) } a \sin ( \omega t ) + o(a^2) $$

Now since

$$ \dot{ \hat{x} }(t) = k a \sin ( \omega t ) \left( J( \hat{x}(t) ) + \frac{ \partial J }{ \partial \hat{x} } \Bigg|_{ \hat{x} = \hat{x}(t) } a \sin ( \omega t )+ o(a^2) \right) $$

is a periodic ( ??? ) time-varying system, averaging may be applied:

$$ \dot{ \hat{x } }(t) = \frac{1}{T} \int_{0}^{T} k a \sin ( \omega t ) \left( J( \hat{x}(t) ) + \frac{ \partial J }{ \partial \hat{x} } \Bigg|_{ \hat{x} = \hat{x}(t) }a \sin ( \omega t ) + o(a^2) \right) dt $$

where $T = \frac{2 \pi}{ \omega }$.

And then:

$$ \frac{1}{T} \int_{0}^{T} k a \sin ( \omega t ) \left( J( \hat{x}(t) ) + \frac{ \partial J }{ \partial \hat{x} } \Bigg|_{ \hat{x} = \hat{x}(t) }a \sin ( \omega t ) + o(a^2) \right) dt = \frac{ka^2}{2} \frac{ \partial J }{ \partial \hat{x} } \Bigg|_{ \hat{x} = \hat{x}(t) } + o(a^4) $$

How was this integral computed?

I can only see a reason if $J$ is periodic as well or even constant in $[0, T]$. Otherwise, I can't see how the integrand is periodic and how $o(a^4)$ appears.

I found this result here on page 7.

Remark: I didn't denote the $ \hat{x} $ in the averaged system with another symbol, but I feel that it would have been more correct to use something like $ \hat{x}_{ \text{av} } $ to avoid ambiguity.

1

There are 1 best solutions below

3
On BEST ANSWER

I looked through the paper you referenced and even the book they reference within $[17]$. I too cannot see why the above is true unless $J$ is periodic or constant. So I think there is a heavy abuse of notation here ("By fixing $\hat{x}(t)$ to a dummy variable $z$") and what they are really saying is as follows:

Instead of using this expression: $$ \dot{ \hat{x} }(t) = k a \sin ( \omega t ) \left( J( \hat{x}(t) ) + \frac{ \partial J }{ \partial \hat{x} } \Bigg|_{ \hat{x} = \hat{x}(t) }a \sin ( \omega t ) + O(a^2) \right) $$

Define $y(s)$ to be the solution of the following differential equation $$ \frac{d}{ds}(y(s)) = k a \sin ( \omega (s+t) ) \left( J( \hat{x}(t) ) + \frac{ \partial J }{ \partial \hat{x} } \Bigg|_{ \hat{x} = \hat{x}(t) }a \sin ( \omega (s+t) ) + O(a^2) \right) $$

Here we have taken the Taylor series in $x$ of $J(x)$, expanded about $\hat{x}(t)$, and evaluated at $\hat{x}(t) + a \sin ( \omega ( s + t ) )$. From that point of view, $y(0) = \hat{x}(t)$ and $y$ as a function of $s$ is periodic with period $T$.

The integration that comes next then makes sense. Note the integral is $O(a^4)$ because $$\int_0^T a \sin (\omega (s + t)) a^k \sin^k(\omega(s + t)) ds$$ is zero iff $k$ is even.