MHB Solution of Periodic ODE with Floquet Theory

Dustinsfl
Messages
2,217
Reaction score
5
For the scalar linear ODE with periodic coefficients,
$$
x' = a(t)x,\quad\quad a(t + T) = a(t),
$$
show that the solution is of the form
$$
x(t) = x_0e^{\mu t}p(t),
$$
where $\mu$ and $x_0$ are constants, and $p(t)$ is a $T$-periodic function.

How can I show the solution is of the form mentioned? Can I just say by Floquet Theory, the solution is of the form $X(t) = \sum\limits_{n=1}^{n}c_nx_n(t)$ where $x_n(t) = e^{\mu_nt}p_n(t)$ but then how do I get the $x_0$ for $c_1$?
By Floquet Theory, we can define $X_k = X(t)v_k$.Suppose $Bv_k = \lambda_kv_k$.
$$
X_k(t + T) = X(t + T)v_k = X(t)Bv_k = X(t)\lambda_kv_k = \lambda_kX(t)v_k = \lambda_kX_k(t)
$$
Let $\underbrace{\lambda_k}_{\text{characteristic multipliers}} = \text{exp}\left[\overbrace{\rho_k}^{\text{characteristic exponents}}T\right]$.
Define $p_k(t) = \frac{X_k(t)}{e^{\rho_k t}}$.
Then
\begin{alignat*}{3}
p_k(t + T) & = & \frac{X_k(t + T)}{\text{exp}\left[\rho_k(t + T)\right]}\\
& = & \frac{\lambda_kX_k(t)}{\text{exp}\left[\rho_k(t + T)\right]}\\
& = & \frac{\lambda_kX_k(t)}{\lambda_k\text{exp}\left[\rho_kt\right]}\\
& = & \frac{X_k(t)}{\text{exp}\left[\rho_kt\right]}\\
& = & p_k(t)
\end{alignat*}
Thus $p(t)$ is T periodic.

Since $x_0 = x(0)$ is an initial condition, can I just say that is why it is constant. What about $\mu$?
 
Last edited:
Physics news on Phys.org
If x is one dimensional then by separation of variables you get:

x(t)=x_0 e^{\int a(t) dt}

Now if you can expand a(t) to power series around 0 (which is not given in your premise), then

a(t) = a(0)+ a'(0)t + a''(0)t^2/2!+...

a(T)=a(0)=\mu

you'll get one part of exp(\mu t) and the other part is
p(t)=e^{a'(0)t^2/2+a''(0)t^3/6+...}

p'(t)=p(t) (a(t)-a(0))=p(t)(a(t)-a(T))

plug t-> t+T to get:
p'(t+T)=0
i.e p(t+T)=const

And this I believe finishes the proof, but it's not general without assuming something on our 'a'.
 
Alan said:
If x is one dimensional then by separation of variables you get:

x(t)=x_0 e^{\int a(t) dt}

Now if you can expand a(t) to power series around 0 (which is not given in your premise), then

a(t) = a(0)+ a'(0)t + a''(0)t^2/2!+...

a(T)=a(0)=\mu

you'll get one part of exp(\mu t) and the other part is
p(t)=e^{a'(0)t^2/2+a''(0)t^3/6+...}

p'(t)=p(t) (a(t)-a(0))=p(t)(a(t)-a(T))

plug t-> t+T to get:
p'(t+T)=0
i.e p(t+T)=const

And this I believe finishes the proof, but it's not general without assuming something on our 'a'.

By separation of variables, we have that $\int\frac{\dot{x}}{x}dx = \int a(t)dt$.
Solving the integral leads too
$$
x = C\exp\left[\int a(t)dt\right].
$$
Let's expand the the Taylor series of $a(t)$ about 0.
$$
a(0) = a(0) + a'(0)t + \frac{a''(0)t^2}{2} + \cdots
$$
Since $a(t)$ is periodic with period $T$, we have $a(T) = a(0) = \mu$.
By substitution, we have
\begin{alignat*}{3}
x(0) & = & C\exp\left[\int \mu dt\right]\\
& = & Ce^{\mu t}
& = & C = x_0
\end{alignat*}
Therefore, we have $x(t) = x_0e^{\mu t}$

But why is p(t) multiplied by x? How do we go from the x above to saying it is $x(t) = x_0e^{\mu t}p(t)$
 
The solution is
x(t)=x_0 e^{a(T)t} e^{a'(0)t^2/2+a''(0)t^3/6+...}

Denote by p(t) the third factor, you can see that it satisfies:

p'(t)=p(t)(a(t)-a(0))
So you get that p'(T)=0, and this is true for any m integer that p'(mT)=0, thus p(t+T)=p(t).
 
Alan said:
The solution is
x(t)=x_0 e^{a(T)t} e^{a'(0)t^2/2+a''(0)t^3/6+...}

Denote by p(t) the third factor, you can see that it satisfies:

p'(t)=p(t)(a(t)-a(0))
So you get that p'(T)=0, and this is true for any m integer that p'(mT)=0, thus p(t+T)=p(t).

How do you go from
$$
\exp\left[a(t + T)\right]
$$
to
$$
e^{a(T)t} e^{a'(0)t^2/2+a''(0)t^3/6+...}
$$
 
You have x(t)=x_0 e^{\int a(t)dt}

I assumed that I can expand a(t) by a power series (I don't see how to show this otherwise), so a(t)=a(0)+a'(0)t+a''(0)t^2/2!+... and a(0)=a(T) from T-periodicity of a(t), plug back to the integral to get what I wrote.

If I can't expand a(t) with a powers series then I am not sure how to solve this.
 
Alan said:
You have x(t)=x_0 e^{\int a(t)dt}

I assumed that I can expand a(t) by a power series (I don't see how to show this otherwise), so a(t)=a(0)+a'(0)t+a''(0)t^2/2!+... and a(0)=a(T) from T-periodicity of a(t), plug back to the integral to get what I wrote.

If I can't expand a(t) with a powers series then I am not sure how to solve this.

So $p(t) = \exp\left[\int\left(a'(0)t + \frac{a''(0)t^2}{2} + \cdots\right)dt\right]$ is defined this way?
 
Alan said:
p'(t)=p(t)(a(t)-a(0))
So you get that p'(T)=0, and this is true for any m integer that p'(mT)=0, thus p(t+T)=p(t).

Ok so I understand p(t) now. Why are you taking the derivative of p(t) to show it is T-periodic?
 
Alan said:
p'(t)=p(t) (a(t)-a(0))=p(t)(a(t)-a(T))

plug t-> t+T to get:
p'(t+T)=0
i.e p(t+T)=const

I don't get this part.
 
  • #10
This is all wrong. Is there another way to do this?
What I wrote original comes from Floquet. Is it enough to just say what I was doing in post 1?
 
  • #11
Hello, I am a phd student in Romania and I have to study Floquet Theory. I need some help with documentation and research. Do you have any? Appreciate your help
 
  • #13
do you think i could apply this floquet theory in economics?
 
  • #14
Raluca said:
do you think i could apply this floquet theory in economics?

This is just the mathematics about it so you should be able to since the ecnomic version is probably built off of the math.
 
Back
Top