3

Clearly, the Green's Theorem proof does not hold as that relies on the specific conditions of complex differentiability. However, Goursat's proof involving the triangle case and building up from there invokes the Cauchy-Riemann equations nowhere, and just requires a basic definition of differentiability that satisfies:

$$f(x+h)-f(x)-f'(x)h=\psi_x(x+h)h$$

Where $\psi_x(x+h)h$ can be made as small as desired.

It also requires some basic topology and geometric properties of triangles, so I believe any $\Bbb R^2$-type space with the Euclidean topology would satisfy Goursat's lemma. Moreover, the use of this lemma to prove the Cauchy Integral Theorem again requires only some basic notion of contour (or line) integrability, and antiderivatives, which the space $\Bbb R^2$ also has. The homotopic invariance theorem relies on properties of compact sets, which exist in the Euclidean topology, and not really any much else.

This leads to believe that if you have a space which is:

  • Complete
  • Has a notion of differentiability satisfying the above
  • Has a notion of antiderivative, Riemann sum and line integral
  • Is two-dimensional with the Euclidean topology
  • Has compact sets and uniform continuity of continuous functions on them

Then you have homotopic invariance of line integrals of differentiable functions. You can't quite have the Cauchy Integral Formula, as $2\pi i$ is not defined without letting $i$ be complex, however similar expressions to do with winding numbers could arise in other spaces.

For example, by my reasoning, we have the above theorems valid in:

  • $\Bbb R^2$
  • The split-complex numbers
  • The dual numbers

Yet these theorems are always touted as important results of complex analysis, implying that I'm missing something here. What about the proofs of any of these theorems is complex-specific?

FShrike
  • 46,840
  • 3
  • 35
  • 94
  • The derivative is a complex linear function (hence given by multiplication)is what forces very strong conditions on the function; in the real plane say, the derivative is a matrix, so one can talk about partial derivatives or more generally directional derivatives only; for example $(x,y)\to x+2y$ is as smooth as you want as a real function, but one cannot get an antiderivative in any meaningful sense as a function, so there is no homotopy invariance of line integrals and the integral on any curve will depend on the curve as doing it on a unit segment vertically or horizontally shows – Conrad Sep 19 '21 at 13:46
  • You could phrase the theorems of complex analysis as theorems of certain special real vector fields instead, but it makes things less elegant most of the time. – Mark S. Sep 19 '21 at 14:47
  • @Conrad I’m not quite sure what you mean by complex linear. The differential operator is linear, sure, but the derivative of a function is almost never a linear function – FShrike Sep 19 '21 at 20:25
  • yes that is the point - the differential operator (or the derivative regarded as a linear operator) is complex linear hence (since $\mathbb C$ is one dimensional over itself), it is multiplication by a constant, which is the derivative at the given point; that is not true when you look at smooth functions from the plane regarded as $\mathbb R^2$ to itself and actually that condition characterizes the complex differentiable functions among the smooth functions in this context; this essentially implies all the good properties via Green theorem and the like – Conrad Sep 20 '21 at 02:08
  • @Conrad This doesn’t rule out over 1-dimensional-over-themselves spaces, such as split complex numbers, right? Additionally, I am struggling to recall where the property $D(\lambda f)=\lambda D(f)$ was used in any of the proofs - perhaps the sections of the proof where we estimate integrals as being less than or equal to the supremum times length of the contour: $\sup|f|\times L(\gamma)\ge|\int_\gamma f|$ – FShrike Sep 20 '21 at 05:47
  • The topology matters too not only the algebraic structure – Conrad Sep 20 '21 at 12:59
  • @Conrad Sorry to bother you, but my understanding of topology is limited. What prevents one from imposing the Euclidean topology of open balls (+ Euclidean metric) onto the split-complex or dual numbers? – FShrike Sep 20 '21 at 16:16
  • 1
    I do not think that Euclidean topology meshes well with the algebraic structure in either case above; what I suggest is to try and write out your definitions (eg what would you consider a continuous function, a differentiable function etc) in either system above and see where you get bogged down before going into deeper stuff; note that neither the split complex numbesr, nor the dual numbers form fields incidentally - up to isomorphism there is only one field of degree two over the reals namely $\mathbb C$; you can write as $(x,y), x+iy, \mathbb R(t)/(t^2+1)$ or whatever but is still unique – Conrad Sep 20 '21 at 16:44
  • I spent a couple afternoons thinking about this once. The notion of analyticity which corresponds the split complex numbers implies that the coefficient functions are solutions of the homogeneous wave equation, as opposed to the Laplace equation. being "split complex analytic" and having non-zero Jacobian does imply that is a local conformal Lorentzian equivalence, which is an analog to a similar statement for complex numbers. Also, split complex analyticity is equivalent to being differentiable and solving the corresponding "CR" equations. – Geoffrey Sangston Dec 05 '23 at 02:57
  • @GeoffreySangston Interesting! It’s never too late to add something useful to a post. Do you know if split complex analyticity has any nice homotopical properties? – FShrike Dec 05 '23 at 03:02

1 Answers1

4

(Note that when I wrote this originally I did not realize that $j$ is used as the adjoined element. I instead denote $j$ by $s$.)


Preliminary

Here $\mathbb{S}$ denotes the set of split complex numbers $x + s y$; $s^2 = 1$, and $x, y$ are real numbers. Let $\overline{x + s y} = x - s y$, and let $Q(z) = z \bar{z}$. $z \in \mathbb{S}$ has a multiplicative inverse if and only if $Q(z) \neq 0$, and it equals $\frac{\bar{z}}{Q(z)}$.

Split-complex Wirtinger derivatives

Let $z$ denote the identity map, and let $x$ and $y$ denote the coordinate projections, so that $z = x + \mathcal{s} y$. Then \begin{align*} \begin{bmatrix} dz \\ d\bar{z} \end{bmatrix} = \begin{bmatrix} 1 & \mathcal{s} \\ 1 & -\mathcal{s} \end{bmatrix} \begin{bmatrix} dx \\ dy \end{bmatrix} \end{align*}

Let $\{ \frac{\partial}{\partial z}, \frac{\partial}{\partial \bar{z}} \}$ be defined so that $\{ dz, d \bar{z} \}$ is its dual basis. Hence there is an invertible matrix $A$ satisfying

\begin{align*} \begin{bmatrix} \frac{\partial}{\partial z} & \frac{\partial}{\partial \bar{z}} \end{bmatrix} = \begin{bmatrix} \frac{\partial}{\partial x} & \frac{\partial}{\partial y} \end{bmatrix} A. \end{align*}

Multiplying these by the previous matrices, on the right, implies \begin{align*} A = \begin{bmatrix} 1 & \mathcal{s} \\ 1 & -\mathcal{s} \end{bmatrix}^{-1} = \frac{1}{2} \begin{bmatrix} 1 & 1 \\ \mathcal{s} & -\mathcal{s} \end{bmatrix}. \end{align*}

Hence $\frac{\partial}{\partial z} = \frac{1}{2} \left( \frac{\partial}{\partial x} + \mathcal{s} \frac{\partial}{\partial y} \right)$ and $\frac{\partial}{\partial \bar{z}} = \frac{1}{2} \left( \frac{\partial}{\partial x} - \mathcal{s} \frac{\partial}{\partial y} \right)$. Notice that the signs of the split-complex Wirtinger derivatives are perhaps easier to remember than for the usual Wirtinger derivatives.

Characterizing the maps satisfying the split complex Cauchy-Riemann equations

Suppose $\frac{\partial f}{\partial \bar{z}} = 0$ over a neighborhood in $\mathbb{S}$. Then \begin{align*} \left( \frac{\partial^2}{\partial x^2} - \frac{\partial^2}{\partial y^2} \right) f^1 + \mathcal{s} \cdot \left( \frac{\partial^2}{\partial x^2} - \frac{\partial^2}{\partial y^2} \right) f^\mathcal{s} = \frac{\partial^2 f}{\partial x^2} - \frac{\partial^2 f}{\partial y^2} = 4 \frac{\partial^2 f}{\partial z \partial \bar{z}} = 0 \end{align*} over the neighborhood. Hence $f^1$ and $f^\mathcal{s}$ are solutions to the homogeneous wave equation. Therefore there exist differentiable functions $A, B, C, D$ such that \begin{align*} f^1(x + \mathcal{s} y) &= A(x - y) + B(x + y), \\\ f^\mathcal{s}(x + \mathcal{s} y) &= C(x - y) + D(x + y). \end{align*}

$ \frac{\partial f}{\partial \bar{z}} = 0 $ implies

\begin{align*} A'(x - y) + B'(x + y) = \frac{\partial f^1}{\partial x} = \frac{\partial f^\mathcal{s}}{\partial y} = -C'(x - y) + D'(x + y) \end{align*} and \begin{align*} -A'(x - y) + B'(x + y) = \frac{\partial f^1}{\partial y} = \frac{\partial f^\mathcal{s}}{\partial x} = C'(x - y) + D'(x + y). \end{align*}

Hence $A' = -C'$ and $B' = D'$. So we conclude there exists a constant $K \in \mathbb{S}$ such that \[f(z) = f(x + \mathcal{s} y) = K + 2 A(x - y) \frac{1 - s}{2} + 2 B(x + y) \frac{1 + s}{2}.\] The split-complex numbers $u = \frac{1-s}{2}$ and $\bar{u} = \frac{1+s}{2}$ determine a null-basis for $\mathbb{S}$. Let $z^u = x - y$ and $z^\bar{u} = x + y$ denote the coefficients of $z$ as expressed in this basis, so $z = z^u u + z^\bar{u} \bar{u}$. So another way to express the previous conclusion is that there exist real-differentiable $\mathbb{R}$-valued functions $F$ and $G$, such that \[ f(z) = F(x - y) u + G(x + y) \bar{u} = F(z^u) u + G(z^\bar{u}) \bar{u}. \]

We can also explicitly show that any real-differentiable function $f$ of this form satisfies $\frac{\partial f}{\partial \bar{z}} = 0$. First note that \[ \frac{\partial}{\partial \bar{z}} = \frac{1}{2} \left( \frac{\partial}{\partial x} - \mathcal{s} \frac{\partial}{\partial y} \right) = \frac{1}{2} \left( (u + \bar{u}) \frac{\partial}{\partial x} - (\bar{u} - u) \frac{\partial}{\partial y} \right) = \frac{u}{2}\left( \frac{\partial}{\partial x} + \frac{\partial}{\partial y}\right) + \frac{\bar{u}}{2} \left( \frac{\partial}{\partial x} - \frac{\partial}{\partial y} \right) \] So, noting that $u^2 = u$, $\bar{u}^2 = \bar{u}$, and $\bar{u} u = 0$, \begin{align*} \frac{\partial f}{\partial \bar{z}} = \left( \frac{u}{2}\left( \frac{\partial}{\partial x} + \frac{\partial}{\partial y}\right) + \frac{\bar{u}}{2} \left( \frac{\partial}{\partial x} - \frac{\partial}{\partial y} \right) \right) \left( F(x - y) u + G(x + y) \bar{u} \right) = 0. \end{align*}

Green's Theorem implies $C^1$ split complex Cauchy theorem

Green's Theorem says (e.g., as in Gamelin) $$\int_{\partial D} P dx + Q dy = \int \int_D \left(\frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y}\right) dx dy,$$ where $D$ denotes a bounded domain such that $\partial D$ is piecewise smooth, and $P$ and $Q$ are $C^1$.

For $C^1$ $f : \mathbb{S} \to \mathbb{S}$ satisfying $\frac{\partial f}{\partial \bar{z}} = 0$, which is equivalent to $f$ simultaneously satisfying $\frac{\partial f^1}{\partial x} = \frac{\partial f^s}{\partial y}$ and $\frac{\partial f^1}{\partial y} = \frac{\partial f^s}{\partial x}$,

\begin{align*} \int_{\partial D} f(z) dz &= \int_{\partial D} f^1(z) dx + f^s(z) dy + s \int_{\partial D} f^s(z) dx + f^1(z) dy \\ &= \int \int_D \frac{\partial f^s}{\partial x} - \frac{\partial f^1}{\partial y} dx dy + \int \int_D \frac{\partial f^1}{\partial x} - \frac{\partial f^s}{\partial y} dx dy \\ &= 0. \end{align*}

Jacobian of split complex analytic map in terms of split complex Wirtinger derivatives

For any (real) differentiable function $f : \mathbb{S} \to \mathbb{S}$, the dual-basis relations show $df$ can be expressed as $df = \frac{\partial f}{\partial z} dz + \frac{\partial f}{\partial \bar{z}} d\bar{z}.$ Label the coordinate projections so that $f = u + \mathcal{s} v$. Then

\begin{align*} \begin{bmatrix} \frac{\partial f}{\partial z} & \frac{\partial f}{\partial \bar{z}} \\\ \frac{\partial \bar{f}}{\partial z} & \frac{\partial \bar{f}}{\partial \bar{z}} \end{bmatrix} \begin{bmatrix} dz \\\ d\bar{z} \end{bmatrix}&= \begin{bmatrix} df \\\ d\bar{f} \end{bmatrix} \\ &= \begin{bmatrix} 1 & \mathcal{s} \\\ 1 & -\mathcal{s} \end{bmatrix}\begin{bmatrix} du \\\ dv \end{bmatrix} \\ &= \begin{bmatrix} 1 & \mathcal{s} \\\ 1 & -\mathcal{s} \end{bmatrix}\begin{bmatrix} \frac{\partial u}{\partial x} & \frac{\partial u}{\partial y} \\\ \frac{\partial v}{\partial x} & \frac{\partial v}{\partial y} \end{bmatrix} \begin{bmatrix} dx \\\ dy \end{bmatrix} \\ &= \begin{bmatrix} 1 & \mathcal{s} \\\ 1 & -\mathcal{s} \end{bmatrix}\begin{bmatrix} \frac{\partial u}{\partial x} & \frac{\partial u}{\partial y} \\\ \frac{\partial v}{\partial x} & \frac{\partial v}{\partial y} \end{bmatrix} \begin{bmatrix} 1 & \mathcal{s} \\\ 1 & -\mathcal{s} \end{bmatrix}^{-1} \begin{bmatrix} dz \\\ d\bar{z} \end{bmatrix}. \end{align*}

This shows that the Jacobian determinant of $f$ equals $Q(\frac{\partial f}{\partial z})$ in the case $\frac{\partial f}{\partial \bar{z}} = 0$.

Local conformal Lorentzian equivalence is equivalent to satisfying the split Cauchy-Riemann equations plus non-vanishing Jacobian (Geometric interpretation of split complex analyticity)

Suppose $f$ satisfies $\frac{\partial f}{\partial \bar{z}} = 0$ on a neighborhood, and also that the Jacobian of $f$ is nonzero. We previously showed that, in the case of split-complex analyticity, this means $Q(\frac{\partial f}{\partial z}) \neq 0$.

Hence \begin{align*} f^\ast \left( Q(dz) \right) = Q(f^\ast dz) = Q(\frac{\partial f}{\partial z} dz + \frac{\partial f}{ d \bar{z}} d\bar{z}) = Q(\frac{\partial f}{\partial z} dz) = Q(\frac{\partial f}{\partial z}) Q(dz). \end{align*}

And so we conclude that $f$ is a local conformal Lorentzian equivalence. (Geometrically, this means that the lightlike geodesics of $Q(dz)$ and $f^\ast \left( Q(dz) \right)$ parameterize the same sets... so $f$ locally shuffles around the lines parallel to $u$, and the lines parallel to $\bar{u}$.)

Conversely, suppose $f^\ast \left( Q(dz) \right) = A \cdot Q(dz)$ for some real-valued function $A$, and also that $Q(\frac{\partial f}{\partial z})$ is non-vanishing. Then

\begin{align*} A \cdot Q(dz) &= Q(f^\ast dz) \\\ &= Q(\frac{\partial f}{\partial z} dz + \frac{\partial f}{ \partial \bar{z}} d\bar{z} ) \\\ &= \left( Q(\frac{\partial f}{\partial z}) + Q(\frac{\partial f}{\partial \bar{z}}) \right) Q(dz) + \frac{\partial f}{\partial z} \overline{\left( \frac{\partial f}{\partial \bar{z}} \right)} dz^2 + \frac{\partial f}{\partial \bar{z}} \overline{\left( \frac{\partial f}{\partial z} \right)} d\bar{z}^2. \end{align*}

Linear independence (over the reals) of $Q(dz)$, $dz^2$, and $d\bar{z}^2$ implies $\frac{\partial f}{\partial z} \overline{\left( \frac{\partial f}{\partial \bar{z}} \right)} = 0$. The values of $\frac{\partial f}{\partial z}$ have multiplicative inverses since $Q(\frac{\partial f}{\partial z})$ is non-vanishing, so this last expression implies $\frac{\partial f}{\partial \bar{z}} = 0$. And so we also find that $A = Q(\frac{\partial f}{\partial z})$.

Split-complex differentiability is equivalent to satisfying the split Cauchy-Riemann equations

Say $f : \mathbb{S} \to \mathbb{S}$ is split-complex differentiable at $a \in \mathbb{S}$ if the limit \[ f'(a) := \lim_{\substack{z \to a \\\ Q(z - a) \neq 0}} \frac{f(z) - f(a)}{z - a} \] exists and is finite. ($Q(z) := z \bar{z}$.) Notice this implies the following limits exist, are finite, and satisfy \[ \frac{\partial f}{\partial x}(a) = \lim_{t \to 0} \frac{f(a + t) - f(a)}{t} = f'(a) = \lim_{t \to 0} \frac{f(a + ts) - f(a)}{ts} = \frac{1}{s} \frac{\partial f}{\partial y}. \]

So $\frac{\partial f}{\partial \bar{z}}(a) = 0$.

Conversely, suppose $f(x, y) = f^1(x, y) + \mathcal{s}\ f^{\mathcal{s}}(x, y)$ is continuously differentiable (as a real map) on a neighborhood $U$ of $a$, and that $\frac{\partial f}{\partial \bar{z}}(a) = 0$. The above characterization states that there exist differentiable real functions $F$ and $G$ such that $f(z) = F(z^u) u + G(z^{\bar{u}}) \bar{u}$. Then we find \begin{align*} \lim_{\substack{z \to a \\\\\\ Q(z - a) \neq 0}} \frac{f(z) - f(a)}{z - a} &= \lim_{\substack{z \to a \\\\\\ z^u \neq a^u, z^\bar{u} \neq a^\bar{u}}} \frac{(F(z^u) - F(a^u)) u + (G(z^{\bar{u}}) - G(a^{\bar{u}})) \bar{u}}{(z^u - a^u) u + (z^{\bar{u}} - a^{\bar{u}}) \bar{u} } \\ &= \left(\lim_{\substack{z \to a \\\\\\ z^u \neq a^u}} \frac{F(z^u) - F(a^u)}{z^u - a^u}\right) u + \left(\lim_{\substack{z \to a \\\\\\ z^\bar{u} \neq a^\bar{u}}} \frac{G(z^\bar{u}) - G(a^\bar{u})}{z^\bar{u} - a^\bar{u}}\right) \bar{u} \\ &= F'(a^u) u + G'(a^\bar{u}) \bar{u}. \end{align*} So the limit exists by differentiability of $F$ and $G$.

  • I didn't ever expect to share this, so it may need to be cleaned up in various ways (like changing $s$ to $j$). I also have one other calculation (which I can share) but this post is rather long. And I recognize this doesn't explicitly answer your question, but I may be able to edit it into a more explicit answer in the future, or perhaps this will just be a post of possibly relevant facts, which is hopefully still useful. – Geoffrey Sangston Dec 05 '23 at 04:04
  • Very impressive. Thanks for your efforts. (I asked this question two years ago but I'm still hopeful someone can comment on homotopy invariance) – FShrike Dec 05 '23 at 13:18
  • @FShrike I remembered that the $C^1$ Cauchy theorem also follows. And I decided to throw in the "Geometric interpretation section" since it's in my notes and I don't want to separate things. Though it's a bit off topic, perhaps, and I can remove it. – Geoffrey Sangston Dec 05 '23 at 15:42
  • No, don't delete anything. These posts are for general appreciation after all, the more information the merrier – FShrike Dec 05 '23 at 16:02
  • 1
    @FShrike I believe I fixed the part with the error. – Geoffrey Sangston Dec 09 '23 at 23:34