Integrating factor and homogeneous equation for SDEs.

99 Views Asked by At

Can someone explain how an integrating factor is obtained when solving SDEs.

An example would be when finding the solution for a general linear SDE: $dX_t = (a(t)X_t +b(t))dt + (c(t)X_t +d(t))dB_t, X_0 = x_0$

My lecturer states the integrating factor to apply to the SDE is $Y_{t}^{-1}$, where $Y_t$ is the solution for $dY_t = a(t)Y_tdt +c(t)Y_tdB_t$. He states this is the homogeneous version of the original SDE.

I understand integrating factors and the homogeneous version in an ODE context, but I'm not sure how these ideas extend to SDEs. Can someone explain these concepts to me. Thanks!

1

There are 1 best solutions below

1
On

Let $\{Y_t\}_{t\in [0,T]}$ be the solution of the homogeneous equation $$dY_t=a(t)Y_tdt+c(t)Y_tdB_t,\quad Y_0=1,$$ i.e. some sort of generalized Geometric Brownian motion which is explicitly given by $$Y_t=\exp\left( \int_0^t c(s)dB_s+\int_0^t [a(s)-c(s)/2]ds\right)$$

Then define $Z_t:=X_tY_t^{-1}$ and use the product rule $$ (\star)\quad dZ_t= X_t d(Y_t^{-1})+Y^{-1}dX_t+d\langle X, Y^{-1}\rangle_t,$$ where the angle brackets denote the quadratic covariation.

First thing to do now is to compute the stochastic differential of $Y_t^{-1}$ using the Itô formula:

$$d(Y_t^{-1})=\frac{-1}{Y_t^2}dY_t+\frac{1}{Y_t^3} d\langle Y\rangle_t,$$

which is the same as

$$d(Y_t^{-1})=\frac{-1}{Y_t^2}\big[a(t)Y_tdt+c(t)Y_tdB_t\big]+\frac{1}{Y_t^3} c(t)^2Y_t^2dt.$$

Notice that some terms cancel out, so after some reordering we obtain

$$d(Y_t^{-1})=\frac{-1}{Y_t}\big[a(t)dt+c(t)dB_t-c(t)^2dt\big].$$

Plugging this into $(\star)$ we obtain \begin{align} dZ_t&= \frac{-X_t}{Y_t}\big[a(t)dt+c(t)dB_t-c(t)^2dt\big]+\frac{1}{Y_t}\big[(a(t)X_t +b(t))dt + (c(t)X_t +d(t))dB_t\big]\\ &-\bigg[\frac{X_t}{Y_t}c(t)^2+\frac{c(t)d(t)}{Y_t}\bigg]dt\\ &= [b(t)- c(t)d(t)]Y_t^{-1}dt+d(t)Y_t^{-1}dB_t. \end{align}

Now integrate to obtain

$$Z_t=X_0+\int_0^t [b(s)-c(s)d(s)]Y_s^{-1}dt+\int_0^t d(s)Y_s^{-1}dB_s,$$ using the fact that $X_t=Z_tY_t$ we obtain

$$X_t=Y_t\left(X_0+\int_0^t [b(s)-c(s)d(s)]Y_s^{-1}dt+\int_0^t d(s)Y_s^{-1}dB_s\right).$$