Sufficient condition for existence and uniqueness of solution of 1st1^{st} order linear ODE


Note: Through out the discussion we assume the following hypothesis:

HIVPH_{IVP}: Let f:I×(ΩRn)(ΩRn)f:\mathbb{I}\times (\Omega \subseteq \mathbb{R}^n) \rightarrow (\Omega \subseteq \mathbb{R}^n) be a continuous function with IR\mathbb{I} \subseteq \mathbb{R} containing x0Rx_{0} \in \mathbb{R}. Here nNn \in \mathbb{N}.

Now let's formalise the condition:

Theorem (Picard-Lindelof):
Consider the following IVP:
dydx=f(x,y)y(x0)=y0\begin{equation} \frac{dy}{dx}=f(x,y) \quad y(x_0)=y_0 \end{equation}
where ff satisfies the HIVPH_{IVP}, where (x0,y0)I×Ω(x_0,y_0) \in \mathbb{I} \times \Omega being interior point of the set along with fy\frac{\partial f}{\partial y} also satisfies HIVPH_{IVP}. Also consider the rectangle:
R={(x,y)I×Ω:xx0a,yy0b,a,bR+}I×ΩR=\{(x,y)\in \mathbb{I} \times \Omega: |x-x_0|\leq a, ||y-y_0||\leq b, \quad a,b \in \mathbb{R}_+ \} \subseteq \mathbb{I}\times \Omega

Let M=SupxRfM=\operatorname{Sup}_{x\in R}f then (1) has a solution in a interval contained in (x0δ,x0+δ)(x_0-\delta,x_0+\delta), where δ=min{a,bM}\delta =\operatorname{min}\{a,\frac{b}{M}\}.

Note: A more general condition can be imposed on f as f being Lipsticz continuous on yy variable instead of the condition fy\frac{\partial f}{\partial y} satisfying HIVPH_{IVP}.

Lipsticz Continous: Let f:XRmYRnf:X\subseteq \mathbb{R}^m \rightarrow Y\subseteq \mathbb{R}^n is called a Lipsticz fucntion if f(x)f(y)nkxym||f(x)-f(y)||_n \leq k ||x-y||_m where kR+k \in \mathbb{R}_+.

*Lipsticz Continous (on yy-variable): Let f:I×(ΩRm)(ΩRm)f:\mathbb{I}\times (\Omega \subseteq \mathbb{R}^m) \rightarrow (\Omega \subseteq \mathbb{R}^m) is called a Lipsticz fucntion if f(x,y1)f(x,y2)mky1y2mfor each fixed xR||f(x,y_1)-f(x,y_2)||_m \leq k ||y_1-y_2||_m \quad \text{for each fixed } x \in \mathbb{R} where kR+k \in \mathbb{R}_+, [where k doesn't depend on x].kk could very well depend on xx in general but or our purpose we enforce the independece of k on x.

Now, if we assume that fy\frac{\partial f}{\partial y} is continous then on the closed bounded set RR in above theorem M0\exists M \geq 0 such that fy(x,y)opM(x,y)R||\frac{\partial f}{\partial y}(x,y)||_{op} \leq M \quad (x,y) \in R. Now, by Lagrange's Mean Value Theorem we get the following inequality:
f(x,y1)f(x,y2)=fy(x,ζ)(y1y2)\begin{equation} f(x,y_1)-f(x,y_2)=\frac{\partial f}{\partial y}(x,\zeta)(y_1-y_2) \end{equation}
where ζ(y1,y2)\zeta \in (y_1,y_2), [where (y1,y2):={y1(1t)+y2t:t(0,1)}(y_1,y_2):=\{y_1(1-t)+y_2t: t\in (0,1)\}]
Now taking norm both side we get,
f(x,y1)f(x,y2)m=fy(x,ζ)(y1y2)m||f(x,y_1)-f(x,y_2)||_m=||\frac{\partial f}{\partial y}(x,\zeta)(y_1-y_2)||_m
    f(x,y1)f(x,y2)m=fy(x,ζ)op(y1y2)mMy1y2m\implies ||f(x,y_1)-f(x,y_2)||_m=||\frac{\partial f}{\partial y}(x,\zeta)||_{op}||(y_1-y_2)||_m \leq M||y_1-y_2||_m
That is f becomes Lipscitz.


Handling nthn^{th} order.

I am here showing a particular 2nd2^{nd}-order case. General case will be similar.

Consider the following IVP:
a0(x)y+a1(x)y+a2(x)y=r(x)y(0)=c1,y(0)=c2\begin{equation} a_0(x)y''+a_1(x)y'+a_2(x)y=r(x) \quad y(0)=c_1,y'(0)=c_2 \end{equation}
where c1,c2Rc_1,c_2 \in \mathbb{R} and (a,b)R(a,b) \subseteq \mathbb{R} such that 0(a,b)0 \in (a,b) and ai(x),r(x)a_i(x),r(x) are continous functions on (a,b)(a,b) with a0(x)0x(a,b)a_0(x)\neq 0 \quad \forall x \in (a,b).

Then (3) has a unique solution in a interval containg 0 and contained in (a,b)(a,b).

Solution:
The trick is to convert the second order ODE to first order ODE and use our known method.

Let Y=(y,y)Y=(y,y')
we have y=ra0a1a0ya2a0yy''=\frac{r}{a_0}-\frac{a_1}{a_0}y'-\frac{a_2}{a_0}y

Then we have,
dYdx=f(x,Y)Y(0)=(y(0),y(0))=(c1,c2)=Y0\begin{equation} \frac{dY}{dx}=f(x,Y) \quad Y(0)=(y(0),y'(0))=(c_1,c_2)=Y_0 \end{equation}
where f(x,Y)=(y,ra0a1a0ya2a0y)f(x,Y)=(y',\frac{r}{a_0}-\frac{a_1}{a_0}y'-\frac{a_2}{a_0}y)

Now, ff is continous function on (a,b)×R2(a,b) \times \mathbb{R}^2. Here we don't even need to consider a subset ΩR2\Omega \subseteq \mathbb{R}^2 containing (c1,c2)(c_1,c_2) as the ff is linear in both y,yy,y' and hence continous on whole of R2\mathbb{R}^2 and a00a_0 \neq 0 in (a,b)(a,b) hence ra0,a1a0,a2a0\frac{r}{a_0},\frac{a_1}{a_0},\frac{a_2}{a_0} are continous on (a,b)(a,b).

Also, fY=(01a2a0a1a0)\frac{\partial f}{\partial Y}=\begin{pmatrix} 0 & 1 \\ -\frac{a_2}{a_0} & -\frac{a_1}{a_0} \end{pmatrix} is also continous on (a,b)×R2(a,b) \times \mathbb{R}^2 for same reason above. Hence by Theorem (Picard-Lindelof) we have a unique solution to the ODE (4)    (3)(4) \implies (3).