MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top


I have the following PDE that I am trying to solve via the Cauchy-Kowalewski Theorem. But I have no idea how to do it or if its possible. Maybe one of you has an idea. Here is the problem: Let $U \subset \mathbb{C}^{n}$ be some open subset witch contains zero. Well you can shrink $U$ arbitralily if you wish. Let $z_{j} = x_{j} + i y_{j}$ be the coordinates on $U$. I am looking for a real-valued analytic function $\beta$ defined on $U$ such that the equations are satisfied: $\frac{\partial \beta}{\partial x_{j}} = F_{j}(x,y,\beta)$ and $\frac{\partial \beta}{\partial y_{j}} = G_{j}(x,y,\beta)$,with $j = 1, ..., n$, where $F,G$ are also real analytic functions and with initial condition $\beta(x,0) = 0$, $\forall x \in U \cap \mathbb{R}^{n}$. Is it possible to solve such an system of equations via Cauchy-Kowalewski ? If yes, how? If no, why and is there any other method that can give me a solution? I hope that for a lot of answers and please excuse me if the question is too trivial. Thanks in advance.

Greeting Andrei

share|cite|improve this question
If you are specifying the partial derivatives, you need necessarily compatibility conditions. As stated the PDE can be overdetermined. Do you know that $\partial_{y_k} F_j = \partial_{x_j} G_k$ for example (they need to be since partial derivatives commute on $\beta$)? Also, typically Cauchy-Kowelewski is used to solve a "initial value problem" where the data is prescribed on a co-dimension 1 set. You need something closer to Cartan-Kahler. – Willie Wong Oct 3 '12 at 7:18
Deane Yang's paper "Local Solvability of Overdetermined Systems Defined by Commuting First-Order Differential Operators" may help (1986, CPAM), I'll see if I can get him to say a few words here. – Willie Wong Oct 3 '12 at 7:19
Thanks that would be very nice! – Andrei Oct 3 '12 at 7:57
@Willie: You have to be a bit careful with the integrability conditions. Since $F$ and $G$ involve the unknown $\beta$, you can't just check whether $\partial_{y^k}F_j=\partial_{x^j}G_k$, since expanding these out will involve the partials of the unknown $\beta$, and you won't know whether you have equality until you know the solution $\beta$ that you are hoping to find, since these equations only have to hold for the solution $\beta$ that also satisfies the given initial conditions (assuming that it actually exists). – Robert Bryant Oct 3 '12 at 14:22
@Robert: can one not plug in the equation there? $$\partial_{y^k}(F_j(x,y,\beta)) = (\partial_{y^k}F_j)(x,y,\beta) + (\partial_\beta F_j)(x,y,\beta)G_j(x,y,\beta)$$ Of course this still depends on the unknown $\beta$, but if the integrability condition is only satisfied for isolated values of $\beta$ one may expect trouble. – Willie Wong Oct 3 '12 at 15:12
up vote 10 down vote accepted

You don't need the Cauchy-Kowalewski Theorem for your problem. In fact, real-analyticity is a red herring here. What you are asking for is a function $\beta(x,y)$ such that the graph $\bigl(x,y,\beta(x,y)\bigr)$ is an integral manifold of the $1$-form $$ \theta = d\beta - F_j(x,y,\beta)\ dx^j - G_j(x,y,\beta)\ dy^j $$ (sum on $j$ in both terms) defined on $\mathbb{R}^{2n+1} = \mathbb{C}^n\times \mathbb{R}$ (or some open neighborhood of $0$ in this space. This makes sense for smooth functions of course, and whether there is a solution or not doesn't depend on real-analyticity.

In fact, you have added the requirement that the $2n$-dimensional graph contain the $n$-dimension submanifold defined by $(x,y,\beta) = (x, 0, 0)$, and you can see from the above that $\theta$ vanishes on that graph if and only if the $F_j$ satisfy $F_j(x,0,0)\equiv0$, so this is certainly a necessary condition.

A sufficient condition, after that, would be, for example, that $d\theta\wedge\theta=0$, for then the Frobenius Theorem would apply. However, this is not necessary if all you are asking is that there be a solution to the specific 'initial value problem' you have posed. To get sufficient conditions, what you should do is use ODE to, for example, construct $\beta_1(x,y^1)$ satisfying the equation $$ \frac{\partial\beta_1}{\partial y^1} = G_1(x,y^1,0,\ldots,0,\beta_1) $$ with the initial condition $\beta_1(x,0) = 0$. Then you need to check that $\theta$ vanishes on the $(n{+}1)$-dimensional graph $\bigl(x,y^1,0,\ldots,0,\beta_1(x,y^1)\bigr)$. Next, you construct $\beta_2(x,y^1,y^2)$ by solving the equation $$ \frac{\partial\beta_2}{\partial y^2} = G_2(x,y^1,y^2,0,\ldots,0,\beta_2) $$ with the initial condition $\beta_2(x,y^1,0) = \beta_1(x,y^1)$, and so on. At each stage, you'll get more conditions on the functions $G_j$ and $F_j$ in order for the constructed graph to be an integral of $\theta$. When you get to the end, these will be the necessary and sufficient conditions for this particular initial value problem.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.