info prev up next book cdrom email home

Ordinary Differential Equation--First-Order Exact

Consider a first-order ODE in the slightly different form

\begin{displaymath}
p(x,y)\,dx+q(x,y)\,dy = 0.
\end{displaymath} (1)

Such an equation is said to be exact if
\begin{displaymath}
{\partial p\over\partial y} = {\partial q\over \partial x}.
\end{displaymath} (2)

This statement is equivalent to the requirement that a Conservative Field exists, so that a scalar potential can be defined. For an exact equation, the solution is
\begin{displaymath}
\int^{(x,y)}_{(x_0,y_0)} p(x,y)\,dx+q(x,y)\,dy = c,
\end{displaymath} (3)

where $c$ is a constant.


A first-order ODE (1) is said to be inexact if

\begin{displaymath}
{\partial p\over \partial y} \not= {\partial q\over\partial x}.
\end{displaymath} (4)

For a nonexact equation, the solution may be obtained by defining an Integrating Factor $\mu$ of (6) so that the new equation
\begin{displaymath}
\mu p(x,y)\,dx+\mu q(x,y)\,dy = 0
\end{displaymath} (5)

satisfies
\begin{displaymath}
{\partial \over \partial y} (\mu p) = {\partial \over \partial x} (\mu q),
\end{displaymath} (6)

or, written out explicitly,
\begin{displaymath}
p{\partial \mu \over \partial y} + \mu {\partial p\over \par...
...tial \mu \over \partial x} + \mu {\partial p\over \partial x}.
\end{displaymath} (7)

This transforms the nonexact equation into an exact one. Solving (7) for $\mu$ gives
\begin{displaymath}
\mu={q{\partial \mu\over \partial x}-p{\partial \mu\over \pa...
... {\partial p
\over \partial y}-{\partial q\over \partial x}}.
\end{displaymath} (8)

Therefore, if a function $\mu$ satisfying (8) can be found, then writing
$\displaystyle P(x,y)$ $\textstyle =$ $\displaystyle \mu p$ (9)
$\displaystyle Q(x,y)$ $\textstyle =$ $\displaystyle \mu q$ (10)

in equation (5) then gives
\begin{displaymath}
P(x,y)\,dx+Q(x,y)\,dy = 0,
\end{displaymath} (11)

which is then an exact ODE. Special cases in which $\mu$ can be found include $x$-dependent, $xy$-dependent, and $y$-dependent integrating factors.


Given an inexact first-order ODE, we can also look for an Integrating Factor $\mu(x)$ so that

\begin{displaymath}
{\partial \mu\over\partial y} = 0.
\end{displaymath} (12)

For the equation to be exact in $\mu p$ and $\mu q$, the equation for a first-order nonexact ODE
\begin{displaymath}
p{\partial\mu\over\partial y} +\mu {\partial p\over\partial ...
...q{\partial\mu\over\partial x} +\mu {\partial p\over\partial x}
\end{displaymath} (13)

becomes
\begin{displaymath}
\mu{\partial p\over\partial y}= q {\partial\mu\over\partial x}+\mu {\partial p\over\partial x}.
\end{displaymath} (14)

Solving for $\partial\mu/\partial x$ gives
\begin{displaymath}
{\partial\mu\over\partial x} =\mu(x) {{\partial p\over\partial y} - {\partial q\over\partial x}\over q}
\equiv f(x,y)\mu(x),
\end{displaymath} (15)

which will be integrable if
\begin{displaymath}
f(x,y)\equiv {{\partial p\over\partial y} - {\partial q\over\partial x}\over q}=f(x),
\end{displaymath} (16)

in which case
\begin{displaymath}
{d\mu\over\mu} = f(x)\,dx,
\end{displaymath} (17)

so that the equation is integrable
\begin{displaymath}
\mu (x) = e^{\int f(x)\,dx},
\end{displaymath} (18)

and the equation
\begin{displaymath}[\mu p(x,y)]dx+[\mu q(x,y)]dy = 0
\end{displaymath} (19)

with known $\mu(x)$ is now exact and can be solved as an exact ODE.


Given in an exact first-order ODE, look for an Integrating Factor $\mu(x,y)=g(xy)$. Then

\begin{displaymath}
{\partial\mu\over\partial x} = {\partial g\over\partial x} y
\end{displaymath} (20)


\begin{displaymath}
{\partial\mu\over\partial y} = {\partial g\over\partial y} x.
\end{displaymath} (21)

Combining these two,
\begin{displaymath}
{\partial\mu\over\partial x} = {y\over x} {\partial\mu\over\partial y}.
\end{displaymath} (22)

For the equation to be exact in $\mu p$ and $\mu q$, the equation for a first-order nonexact ODE
\begin{displaymath}
p{\partial\mu\over\partial y} +\mu {\partial p\over\partial ...
...q{\partial\mu\over\partial x} +\mu {\partial p\over\partial x}
\end{displaymath} (23)

becomes
\begin{displaymath}
{\partial\mu\over\partial y}\left({p - {y\over x}q}\right)
...
...al p\over\partial x} - {\partial p\over\partial y}}\right)\mu.
\end{displaymath} (24)

Therefore,
\begin{displaymath}
{1\over x} {\partial\mu\over\partial y}
= {{\partial q\over\partial x} - {\partial p\over\partial y}\over xp-yq}\mu.
\end{displaymath} (25)

Define a new variable
\begin{displaymath}
t(x,y)\equiv xy,
\end{displaymath} (26)

then ${\partial t/\partial y} = x$, so
\begin{displaymath}
{\partial\mu\over\partial t} = {\partial\mu\over\partial y}{...
...ial p\over\partial y}\over xp-yq}\mu(t)
\equiv f(x,y)\mu (t).
\end{displaymath} (27)

Now, if
\begin{displaymath}
f(x,y)\equiv {{\partial q\over\partial x} - {\partial p\over\partial y}\over xp-yq} = f(xy) = f(t),
\end{displaymath} (28)

then
\begin{displaymath}
{\partial\mu\over\partial t} = f(t)\mu (t),
\end{displaymath} (29)

so that
\begin{displaymath}
\mu = e^{\int f(t)\,dt}
\end{displaymath} (30)

and the equation
\begin{displaymath}[\mu p(x,y)]\,dx+[\mu q(x,y)]\,dy = 0
\end{displaymath} (31)

is now exact and can be solved as an exact ODE.


Given an inexact first-order ODE, assume there exists an integrating factor

\begin{displaymath}
\mu = f(y),
\end{displaymath} (32)

so $\partial\mu/\partial x = 0$. For the equation to be exact in $\mu p$ and $\mu q$, equation (7) becomes
\begin{displaymath}
{\partial\mu\over\partial y} = {{\partial q\over\partial x} - {\partial p\over\partial y}\over p}\mu = f(x,y)\mu (y).
\end{displaymath} (33)

Now, if
\begin{displaymath}
f(x,y)\equiv {{\partial q\over\partial x} - {\partial p\over\partial y}\over p} = f(y),
\end{displaymath} (34)

then
\begin{displaymath}
{d\mu\over\mu} = f(y)\,dy,
\end{displaymath} (35)

so that
\begin{displaymath}
\mu (y) = e^{\int f(y)\,dy},
\end{displaymath} (36)

and the equation
\begin{displaymath}
\mu p(x,y)\,dx+\mu q(x,y)\,dy = 0
\end{displaymath} (37)

is now exact and can be solved as an exact ODE.


Given a first-order ODE of the form

\begin{displaymath}
yf(xy)\,dx + xg(xy)\,dy = 0,
\end{displaymath} (38)

define
\begin{displaymath}
v \equiv xy.
\end{displaymath} (39)

Then the solution is
\begin{displaymath}
\cases{
\ln x = \int{g(v)\,dv\over c[g(v)-f(v)]} + c & for $g(v) \not = f(v)$\cr
xy = c & for $g(v) = f(v)$.\cr}
\end{displaymath} (40)


If

\begin{displaymath}
{dy\over dx} = F(x,y) = G(v),
\end{displaymath} (41)

where
\begin{displaymath}
v \equiv {y\over x},
\end{displaymath} (42)

then letting
\begin{displaymath}
y \equiv xv
\end{displaymath} (43)

gives
\begin{displaymath}
{dy\over dx} =x {dv/dx} + v
\end{displaymath} (44)


\begin{displaymath}
x {dv\over dx} + v = G(v).
\end{displaymath} (45)

This can be integrated by quadratures, so
\begin{displaymath}
\ln x = \int {dv\over f(v)-v} + c \qquad {\rm for\ } f(v) \not = v
\end{displaymath} (46)


\begin{displaymath}
y = cx \qquad {\rm for} f(v) = v.
\end{displaymath} (47)


References

Boyce, W. E. and DiPrima, R. C. Elementary Differential Equations and Boundary Value Problems, 4th ed. New York: Wiley, 1986.



info prev up next book cdrom email home

© 1996-9 Eric W. Weisstein
1999-05-26