2.1 Analytic Functions and the Cauchy–Riemann Equations
What it means for a function of a complex variable to have a derivative, the Cauchy–Riemann equations that characterise it, and the harmonic functions that come with it — together with the counterexamples that show exactly which hypotheses matter.
A function $f:\mathbb{C}\to\mathbb{C}$ is, from one point of view, just a map of the plane to itself, $(x,y)\mapsto(u(x,y),v(x,y))$, and real analysis already tells us what it means for such a map to be differentiable. Complex analysis asks for much more. We require the difference quotient $\dfrac{f(z_0+h)-f(z_0)}{h}$ — a quotient of two complex numbers — to have a limit as $h\to0$ from every direction at once. That single demand forces the two real functions $u$ and $v$ to be tied together by a pair of partial differential equations, the Cauchy–Riemann equations $u_x=v_y$, $u_y=-v_x$, and it has astonishing consequences: a complex-differentiable function is automatically infinitely differentiable (§2.2), equal to its Taylor series (Chapter 3), and determined everywhere by its values on a tiny segment (§2.3). Nothing like this is true for real functions.
This section builds the theory from the definition: limits and continuity, the complex derivative, the Cauchy–Riemann equations as a necessary condition and — with continuity of the partial derivatives — a sufficient one, their polar form, and the rigidity results ("if $|f|$ is constant then $f$ is constant"). The real and imaginary parts of an analytic function turn out to be harmonic, and we learn to reverse the process: given a harmonic $u$, find $v$ (a harmonic conjugate) and the analytic $f=u+iv$, by integrating the Cauchy–Riemann equations or by the Milne–Thomson substitution. CSIR NET tests this material constantly: "at which points is $f$ differentiable?", "find $f$ given $\operatorname{Re}f$", "which of the following functions are harmonic?", and Part C statements whose wrong options are the classical counterexamples collected in the ⚠️ box below.
Limits, Continuity and Domains
We write $z=x+iy$ and, for a function $f$ of $z$, $f=u+iv$ with $u=\operatorname{Re}f$ and $v=\operatorname{Im}f$ regarded as real functions of the two real variables $(x,y)$. The open disc of centre $a$ and radius $r$ is $D(a,r)=\{z:|z-a|\lt r\}$, and $D'(a,r)=D(a,r)\setminus\{a\}$ is the punctured disc.
A set $U\subseteq\mathbb{C}$ is open if every point $a\in U$ has a disc $D(a,r)\subseteq U$. An open set $U$ is connected if it is not the union of two disjoint nonempty open sets; for open subsets of $\mathbb{C}$ this is equivalent to being polygonally connected: any two points of $U$ can be joined by a polygonal path (finitely many line segments) lying in $U$, and the segments may even be taken horizontal and vertical.
A domain (or region) is a nonempty connected open set. Discs, half-planes, annuli $\{r\lt|z-a|\lt R\}$, the punctured plane $\mathbb{C}\setminus\{0\}$ and the cut plane $\mathbb{C}\setminus(-\infty,0]$ are domains; the union of two disjoint discs is open but not a domain.
Let $f$ be defined on a set $S$ and let $z_0$ be a limit point of $S$. We write $\lim_{z\to z_0}f(z)=L$ if for every $\varepsilon\gt0$ there is $\delta\gt0$ such that $|f(z)-L|\lt\varepsilon$ whenever $z\in S$ and $0\lt|z-z_0|\lt\delta$. The function $f$ is continuous at $z_0\in S$ if $\lim_{z\to z_0}f(z)=f(z_0)$ (or $z_0$ is an isolated point of $S$), and continuous on $S$ if it is continuous at every point of $S$.
Limits involving $\infty$ are defined through the substitution $z=1/w$: $\lim_{z\to\infty}f(z)=L$ means $\lim_{w\to0}f(1/w)=L$, and $\lim_{z\to z_0}f(z)=\infty$ means $\lim_{z\to z_0}1/f(z)=0$.
The crucial point is that $z$ may approach $z_0$ in any manner — along any line, spiral or sequence of points. A limit exists only if all these approaches give the same answer.
- If $f=u+iv$ and $L=a+ib$, then $\lim_{z\to z_0}f(z)=L$ iff $u(x,y)\to a$ and $v(x,y)\to b$ as $(x,y)\to(x_0,y_0)$.
- If $\lim f=L$ and $\lim g=M$, then $\lim(f\pm g)=L\pm M$, $\lim fg=LM$, and $\lim f/g=L/M$ provided $M\ne0$.
- Sums, products, quotients (where the denominator is nonzero) and compositions of continuous functions are continuous. Polynomials, $e^z$, $\sin z$, $\cos z$ are continuous on $\mathbb{C}$; if $f$ is continuous, so are $\bar f$, $|f|$, $\operatorname{Re}f$ and $\operatorname{Im}f$.
- If $z_0$ is approached along two paths in $S$ on which $f$ has different limits, then $\lim_{z\to z_0}f(z)$ does not exist.
Proof. (1) follows from $$\max(|\operatorname{Re}w|,|\operatorname{Im}w|)\le|w|\le|\operatorname{Re}w|+|\operatorname{Im}w|$$ with $w=f(z)-L$. (2) For the product, $$|fg-LM|=|(f-L)g+L(g-M)|\le|f-L|\,|g|+|L|\,|g-M|;$$ near $z_0$, $|g|\le|M|+1$, and both terms become small. For the quotient it suffices to treat $1/g$: near $z_0$, $|g|\ge|M|/2$, so $\left|\frac1g-\frac1M\right|=\frac{|g-M|}{|g|\,|M|}\le\frac{2|g-M|}{|M|^2}$. Sums are immediate. (3) The algebraic statements follow from (2). For a composition $g\circ f$ with $f$ continuous at $z_0$ and $g$ continuous at $f(z_0)$: given $\varepsilon$, choose $\eta$ with $|g(w)-g(f(z_0))|\lt\varepsilon$ for $|w-f(z_0)|\lt\eta$, then $\delta$ with $|f(z)-f(z_0)|\lt\eta$ for $|z-z_0|\lt\delta$. Power series are continuous inside their disc of convergence because they are differentiable there (Theorem 1.23, together with Theorem 2.2(2) below). Finally $|\bar f(z)-\bar f(z_0)|=|f(z)-f(z_0)|$ and $\big||f(z)|-|f(z_0)|\big|\le|f(z)-f(z_0)|$. (4) If the limit $L$ existed, every path would give $L$. ∎
The Complex Derivative
Formally this is the definition from real calculus. The difference is that $h$ ranges over a disc, not an interval, and this makes complex differentiability a far stronger property.
- Linear approximation. $f$ is differentiable at $z_0$ iff there is a function $\varphi$ on $U$, continuous at $z_0$, with $f(z)-f(z_0)=\varphi(z)(z-z_0)$ for all $z\in U$; then $f'(z_0)=\varphi(z_0)$. Equivalently, $f(z_0+h)=f(z_0)+f'(z_0)h+o(|h|)$.
- If $f$ is differentiable at $z_0$, it is continuous at $z_0$.
- If $f,g$ are differentiable at $z_0$, so are $f\pm g$, $fg$ and (if $g(z_0)\ne0$) $f/g$, with the usual formulas $(fg)'=f'g+fg'$, $(f/g)'=(f'g-fg')/g^2$.
- Chain rule. If $f$ is differentiable at $z_0$ and $g$ is differentiable at $f(z_0)$, then $g\circ f$ is differentiable at $z_0$ and $(g\circ f)'(z_0)=g'(f(z_0))\,f'(z_0)$.
- $(z^n)'=nz^{n-1}$ for $n\in\mathbb{Z}$ ($z\ne0$ if $n\lt0$). Polynomials are differentiable on $\mathbb{C}$, rational functions at every point where the denominator is nonzero, and power series inside their disc of convergence (Theorem 1.23).
Proof. (1) If $f'(z_0)$ exists, put $\varphi(z)=\frac{f(z)-f(z_0)}{z-z_0}$ for $z\ne z_0$ and $\varphi(z_0)=f'(z_0)$; continuity of $\varphi$ at $z_0$ is exactly the existence of the limit. Conversely, such a $\varphi$ gives $\frac{f(z)-f(z_0)}{z-z_0}=\varphi(z)\to\varphi(z_0)$. (2) $f(z)-f(z_0)=\varphi(z)(z-z_0)\to\varphi(z_0)\cdot0=0$.
(3) Write $f(z)-f(z_0)=\varphi(z)(z-z_0)$ and $g(z)-g(z_0)=\psi(z)(z-z_0)$. Then $$f(z)g(z)-f(z_0)g(z_0)=\big[\varphi(z)g(z)+f(z_0)\psi(z)\big](z-z_0),$$ and the bracket is continuous at $z_0$ with value $f'(z_0)g(z_0)+f(z_0)g'(z_0)$. Similarly $\frac1{g(z)}-\frac1{g(z_0)}=\frac{-\psi(z)}{g(z)g(z_0)}(z-z_0)$ near $z_0$ (where $g\ne0$ by continuity). This is the rule for the derivative of $1/g$, namely $-g'/g^2$; the quotient rule follows from the product rule.
(4) Write $g(w)-g(w_0)=\chi(w)(w-w_0)$ with $w_0=f(z_0)$ and $\chi$ continuous at $w_0$. Then $$g(f(z))-g(f(z_0))=\chi(f(z))\,\varphi(z)\,(z-z_0),$$ and $\chi\circ f$ is continuous at $z_0$ by (2) and Theorem 2.1(3). So $(g\circ f)'(z_0)=\chi(w_0)\varphi(z_0)=g'(w_0)f'(z_0)$. (Writing $\varphi$ instead of the difference quotient avoids the usual problem of dividing by $f(z)-f(z_0)$, which may vanish.)
(5) For $n\ge1$, $$z^n-z_0^n=(z^{n-1}+z^{n-2}z_0+\cdots+z_0^{n-1})(z-z_0),$$ and the bracket tends to $nz_0^{n-1}$. For $n\lt0$ use the quotient rule on $1/z^{|n|}$. The rest follows from (3). ∎
The Cauchy–Riemann Equations
Approach $z_0$ along the two coordinate directions. Along real $h$ the difference quotient becomes the partial derivative $f_x=u_x+iv_x$; along imaginary $h=it$ it becomes $\frac{1}{i}f_y$. If $f'(z_0)$ exists these must agree.
Proof. Take $h=t$ real: $\dfrac{f(z_0+t)-f(z_0)}{t}\to f'(z_0)$, so the partial derivative $f_x(z_0)=u_x+iv_x$ exists and equals $f'(z_0)$ (real and imaginary parts converge separately, Theorem 2.1(1)). Take $h=it$: $\dfrac{f(z_0+it)-f(z_0)}{it}\to f'(z_0)$, so $f_y(z_0)=u_y+iv_y$ exists and equals $if'(z_0)$. Hence $f_y=if_x$, that is, $$u_y+iv_y=i(u_x+iv_x)=-v_x+iu_x.$$ Comparing real and imaginary parts gives $u_y=-v_x$ and $v_y=u_x$. Then $f'(z_0)=u_x+iv_x$, and also $f'=-if_y=v_y-iu_y$. Finally $u_xv_y-u_yv_x=u_x^2+v_x^2=|f'(z_0)|^2$. ∎
The complex form. Define the Wirtinger derivatives $$\frac{\partial f}{\partial z}=\frac12\left(f_x-if_y\right),\qquad\frac{\partial f}{\partial\bar z}=\frac12\left(f_x+if_y\right).$$ Since $f_x+if_y=(u_x-v_y)+i(v_x+u_y)$, the CR equations say exactly $\dfrac{\partial f}{\partial\bar z}=0$, and then $f'=\dfrac{\partial f}{\partial z}$. One checks $\frac{\partial z}{\partial\bar z}=0$, $\frac{\partial\bar z}{\partial\bar z}=1$, and the usual product and chain rules hold for these operators. This gives a fast practical test: substitute $x=\frac{z+\bar z}2$, $y=\frac{z-\bar z}{2i}$ to write $f$ as an expression in $z$ and $\bar z$, and differentiate formally with respect to $\bar z$, treating $z$ as a constant. For instance $|z|^2=z\bar z$ has $\partial/\partial\bar z=z$, which vanishes only at $0$ — the answer of Example 2(c).
The CR equations are necessary; they are not sufficient on their own (Example 3 below). What is missing is the real differentiability of $u$ and $v$.
- $f$ is complex differentiable at $z_0$ if and only if $u$ and $v$ are differentiable at $(x_0,y_0)$ as functions of two real variables (that is, with $h=s+it$, $$u(x_0+s,y_0+t)-u(x_0,y_0)=u_xs+u_yt+o(|h|),$$ and likewise for $v$) and the CR equations hold at $(x_0,y_0)$.
- In particular, if the partial derivatives $u_x,u_y,v_x,v_y$ exist in a neighbourhood of $z_0$, are continuous at $z_0$, and satisfy the CR equations at $z_0$, then $f$ is differentiable at $z_0$, with $f'(z_0)=u_x+iv_x$.
- If $u,v$ have continuous first partial derivatives on $U$ and satisfy the CR equations at every point of $U$, then $f$ is differentiable at every point of $U$.
Proof. (1, ⇒) Let $f'(z_0)=a+ib$ and $h=s+it$. By Theorem 2.2(1), $$f(z_0+h)-f(z_0)=(a+ib)(s+it)+o(|h|)$$ $$=(as-bt)+i(bs+at)+o(|h|).$$ Taking real and imaginary parts, $$u(x_0+s,y_0+t)-u(x_0,y_0)=as-bt+o(|h|),\qquad v(x_0+s,y_0+t)-v(x_0,y_0)=bs+at+o(|h|).$$ So $u$ and $v$ are real-differentiable with $(u_x,u_y)=(a,-b)$ and $(v_x,v_y)=(b,a)$, which satisfy the CR equations.
(1, ⇐) Put $a=u_x=v_y$ and $b=v_x=-u_y$ at $(x_0,y_0)$. Real differentiability gives $$f(z_0+h)-f(z_0)=(as-bt)+i(bs+at)+o(|h|)=(a+ib)h+o(|h|),$$ so $$\frac{f(z_0+h)-f(z_0)}{h}=(a+ib)+\frac{o(|h|)}{h}\to a+ib.$$
(2) It suffices to show that continuity of the partials at the point gives real differentiability. For $u$, split the increment and apply the one-variable mean value theorem twice: $$u(x_0+s,y_0+t)-u(x_0,y_0)=u_x(x_0+\theta_1s,\,y_0+t)\,s+u_y(x_0,\,y_0+\theta_2t)\,t$$ with $\theta_1,\theta_2\in(0,1)$. By continuity at $(x_0,y_0)$, $u_x(x_0+\theta_1s,y_0+t)=u_x(x_0,y_0)+\varepsilon_1$ and $u_y(x_0,y_0+\theta_2t)=u_y(x_0,y_0)+\varepsilon_2$ with $\varepsilon_1,\varepsilon_2\to0$ as $h\to0$, and $|\varepsilon_1s+\varepsilon_2t|\le(|\varepsilon_1|+|\varepsilon_2|)|h|=o(|h|)$. The same for $v$; now apply (1). (3) Apply (2) at each point. ∎
(b) Let $f(z)=\dfrac{x^3+iy^3}{x^2+y^2}$ for $z\ne0$ and $f(0)=0$. Show that $f$ is continuous, satisfies the CR equations at $0$, and is not differentiable at $0$.
In polar coordinates $z=re^{i\theta}$ ($r\gt0$) it is often more natural to write $f(re^{i\theta})=u(r,\theta)+iv(r,\theta)$, for instance for logarithms and powers.
Proof. With $x=r\cos\theta$, $y=r\sin\theta$, the chain rule gives $$u_r=u_x\cos\theta+u_y\sin\theta,\qquad u_\theta=-ru_x\sin\theta+ru_y\cos\theta,$$ and the same for $v$. Since the Jacobian $r$ of $(r,\theta)\mapsto(x,y)$ is nonzero, these can be inverted: $u_x=\cos\theta\,u_r-\frac{\sin\theta}{r}u_\theta$, $u_y=\sin\theta\,u_r+\frac{\cos\theta}{r}u_\theta$, and similarly for $v$; in particular the Cartesian partials are continuous. If the Cartesian CR equations hold, then $$v_\theta=-rv_x\sin\theta+rv_y\cos\theta=ru_y\sin\theta+ru_x\cos\theta=ru_r,$$ $$v_r=v_x\cos\theta+v_y\sin\theta=-u_y\cos\theta+u_x\sin\theta=-\tfrac1r u_\theta.$$ Conversely, if the polar equations hold, substitute $v_\theta=ru_r$ and $v_r=-u_\theta/r$ into the inverted formulas: $$v_y=\sin\theta\,v_r+\tfrac{\cos\theta}{r}v_\theta=-\tfrac{\sin\theta}{r}u_\theta+\cos\theta\,u_r=u_x,$$ $$v_x=\cos\theta\,v_r-\tfrac{\sin\theta}r v_\theta=-\tfrac{\cos\theta}{r}u_\theta-\sin\theta\,u_r=-u_y.$$ Theorem 2.4(3) now applies. For the derivative, $$f'=u_x+iv_x=\cos\theta\,(u_r+iv_r)-\tfrac{\sin\theta}{r}(u_\theta+iv_\theta),$$ and the polar equations give $u_\theta+iv_\theta=-rv_r+iru_r=ir(u_r+iv_r)$. Hence $$f'=(\cos\theta-i\sin\theta)(u_r+iv_r)=e^{-i\theta}(u_r+iv_r).$$ ∎
Applications. (i) On the cut plane $\mathbb{C}\setminus(-\infty,0]$, $\operatorname{Log}z=\ln r+i\theta$ with $-\pi\lt\theta\lt\pi$: $u_r=\frac1r=\frac1rv_\theta$ and $v_r=0=-\frac1ru_\theta$, so $\operatorname{Log}$ is analytic with $(\operatorname{Log}z)'=e^{-i\theta}\cdot\frac1r=\frac1z$ — a second proof of Theorem 1.27(2). (ii) For real $\alpha$, the principal power $z^\alpha=r^\alpha e^{i\alpha\theta}$ has $u=r^\alpha\cos\alpha\theta$, $v=r^\alpha\sin\alpha\theta$; then $u_r=\alpha r^{\alpha-1}\cos\alpha\theta=\frac1rv_\theta$ and $v_r=\alpha r^{\alpha-1}\sin\alpha\theta=-\frac1ru_\theta$, so $z^\alpha$ is analytic on the cut plane with derivative $e^{-i\theta}\alpha r^{\alpha-1}e^{i\alpha\theta}=\alpha z^{\alpha}/z$. (iii) For $\bar z=re^{-i\theta}$: $u_r=\cos\theta$ but $\frac1rv_\theta=-\cos\theta$, and $v_r=-\sin\theta$ but $-\frac1ru_\theta=\sin\theta$; the two equations would need $\cos\theta=\sin\theta=0$, so $\bar z$ is differentiable nowhere — as in Example 2.
Analytic Functions and Their Rigidity
A function $f$ is analytic at $z_0$ if it is complex differentiable at every point of some disc $D(z_0,r)$, $r\gt0$. It is analytic on an open set $U$ if it is differentiable at every point of $U$ (equivalently, analytic at every point of $U$). Holomorphic is a synonym for analytic. A function analytic on all of $\mathbb{C}$ is entire.
A point $z_0$ at which $f$ is not analytic, but such that every disc $D(z_0,r)$ contains points where $f$ is analytic, is a singular point (singularity) of $f$. For example $0$ is a singular point of $1/z$, and every point of $(-\infty,0]$ is a singular point of $\operatorname{Log}z$.
"Analytic at a point" is a condition on a whole neighbourhood. So $|z|^2$, differentiable at $0$ only, is analytic nowhere; $x^2+iy^2$, differentiable on a line, is analytic nowhere. Polynomials, $e^z$, $\sin z$, $\cos z$, $\sinh z$, $\cosh z$ are entire; rational functions are analytic off the zeros of the denominator; $\operatorname{Log}z$ and $z^\alpha$ are analytic on the cut plane. (The name "analytic" — locally a sum of a power series — is justified in Chapter 3, where we prove that every function differentiable on a disc is the sum of its Taylor series there.)
The CR equations make analytic functions remarkably rigid: the real part alone, or the modulus alone, cannot vary without the whole function varying in a prescribed way.
- If $f'(z)=0$ for all $z\in D$, then $f$ is constant on $D$.
- If any one of the following holds on $D$, then $f$ is constant: (i) $\operatorname{Re}f$ is constant; (ii) $\operatorname{Im}f$ is constant; (iii) $|f|$ is constant; (iv) $f(D)$ is contained in a straight line; (v) $\bar f$ is also analytic on $D$.
Proof. (1) By Theorem 2.3, $0=f'=u_x+iv_x=v_y-iu_y$, so all four partial derivatives of $u$ and $v$ vanish on $D$. On any horizontal segment in $D$, $u$ and $v$ are functions of $x$ alone with zero derivative, hence constant (one-variable mean value theorem); likewise on vertical segments. Since $D$ is a domain, any two points of $D$ are joined by a path in $D$ made of horizontal and vertical segments (Definition 2.1), so $u$ and $v$ take the same values at both points.
(2)(i) If $u$ is constant then $u_x=u_y=0$, and by CR $v_y=u_x=0$, $v_x=-u_y=0$. So $f'=u_x+iv_x=0$ and (1) applies. (ii) is the same with the roles exchanged. (iii) If $|f|=0$ we are done. Otherwise $u^2+v^2=c^2$ with $c\gt0$; differentiating, $uu_x+vv_x=0$ and $uu_y+vv_y=0$. Substitute the CR equations $u_y=-v_x$, $v_y=u_x$ into the second: $$uu_x+vv_x=0,\qquad vu_x-uv_x=0.$$ This is a linear system for $(u_x,v_x)$ with determinant $-u^2-v^2=-c^2\ne0$, so $u_x=v_x=0$, $f'=0$, and (1) applies. (iv) A line is a set $\{w:\operatorname{Re}(\beta w)=c\}$ with $\beta\ne0$ and $c$ real; then $g=\beta f$ is analytic with constant real part, so $g$ and hence $f$ is constant by (i). (v) If $\bar f$ is analytic, so is $f+\bar f=2u$, an analytic function with imaginary part $0$; by (ii) it is constant, so $u$ is constant and (i) applies. ∎
(b) Prove: if $f$ is analytic on a domain $D$ and $f(D)$ lies on the circle $|w-1|=2$, then $f$ is constant.
(c) Disprove: "if $f$ is analytic on an open set $U$ and $|f|$ is constant on $U$, then $f$ is constant."
Harmonic Functions and Harmonic Conjugates
- $u$ and $v$ are harmonic on $U$; $v$ is a harmonic conjugate of $u$, and $-u$ is a harmonic conjugate of $v$.
- $\nabla u\cdot\nabla v=0$ and $|\nabla u|=|\nabla v|=|f'|$ at every point. Hence at a point where $f'\ne0$, the level curves $u=c_1$ and $v=c_2$ through that point cross at right angles.
Proof. We use a fact proved in §2.2 (Theorem 2.18): the derivative of an analytic function is again analytic. Applying it repeatedly, $f$ has derivatives of all orders, so $u$ and $v$ have continuous partial derivatives of all orders (by Theorem 2.3 applied to $f',f'',\dots$, e.g. $f'=u_x+iv_x$ and $f''=u_{xx}+iv_{xx}$). In particular mixed partials commute. Then, by CR, $$u_{xx}=(v_y)_x=(v_x)_y=(-u_y)_y=-u_{yy},$$ so $\Delta u=0$; similarly $v_{xx}=(-u_y)_x=-(u_x)_y=-(v_y)_y=-v_{yy}$. Since $-if=v-iu$ is analytic, $-u$ is a harmonic conjugate of $v$. For (2): $\nabla u\cdot\nabla v=u_xv_x+u_yv_y=u_xv_x-v_xu_x=0$, and $|\nabla v|^2=v_x^2+v_y^2=u_y^2+u_x^2=|\nabla u|^2=|f'|^2$. The gradients are normal to the level curves, so when they are nonzero and perpendicular, the level curves are perpendicular. ∎
Note the asymmetry: if $v$ is a harmonic conjugate of $u$, then $u$ is a harmonic conjugate of $v$ only when both are constant — the correct statement is that $-u$ is a conjugate of $v$. (If $u+iv$ and $v+iu$ were both analytic, so would be their difference $(u-v)(1-i)$, which has constant argument, hence is constant by Theorem 2.6.)
Conversely, given a harmonic $u$, can we find $v$? Locally always; globally only if the domain has no holes.
- If $u$ is harmonic on a disc $D$ (or on all of $\mathbb{C}$), then $u$ has a harmonic conjugate on $D$. The same holds on every simply connected domain (proved in §2.2, after Theorem 2.15).
- On a domain, any two harmonic conjugates of $u$ differ by a real constant.
- On a domain that is not simply connected a harmonic conjugate need not exist: $u=\ln|z|$ is harmonic on $\mathbb{C}\setminus\{0\}$ but has no harmonic conjugate there.
Proof. (1) Let $(x_0,y_0)$ be the centre of $D$. For $(x,y)\in D$ the path from $(x_0,y_0)$ horizontally to $(x,y_0)$ and then vertically to $(x,y)$ stays in $D$ (each point of it is at least as close to the centre as $(x,y)$). Define $$v(x,y)=\int_{y_0}^{y}u_x(x,s)\,ds-\int_{x_0}^{x}u_y(t,y_0)\,dt.$$ By the fundamental theorem of calculus, $v_y(x,y)=u_x(x,y)$. Differentiating under the integral sign (legitimate because $u_{xx}$ is continuous), $$v_x(x,y)=\int_{y_0}^{y}u_{xx}(x,s)\,ds-u_y(x,y_0)=-\int_{y_0}^{y}u_{yy}(x,s)\,ds-u_y(x,y_0)=-u_y(x,y).$$ So $u,v$ have continuous first partials satisfying CR, $f=u+iv$ is analytic by Theorem 2.4(3), and $v$ is harmonic by Theorem 2.7. On $\mathbb{C}$ the same formula works for every $(x,y)$.
(2) If $v_1,v_2$ are harmonic conjugates of $u$, then $(u+iv_1)-(u+iv_2)=i(v_1-v_2)$ is analytic with real part $0$, hence constant (Theorem 2.6).
(3) $\ln|z|=\frac12\ln(x^2+y^2)$ is harmonic (it is $\operatorname{Re}\operatorname{Log}z$ near each point off the cut, and a direct computation works everywhere: $u_{xx}=\frac{y^2-x^2}{(x^2+y^2)^2}=-u_{yy}$). Suppose $v$ were a harmonic conjugate on $\mathbb{C}\setminus\{0\}$. On the cut plane, $\operatorname{Arg}z$ is also a harmonic conjugate of $\ln|z|$, so by (2), $v=\operatorname{Arg}z+c$ there. Then as $z$ approaches a point $x\lt0$ from above, $v\to\pi+c$, and from below $v\to-\pi+c$, contradicting the continuity of $v$ at $x$. ∎
- Check first that $u$ is harmonic. If $\Delta u\ne0$, no analytic $f$ has real part $u$ — whatever the problem seems to suggest.
- Recognise. Many harmonic functions are real parts of familiar functions: $x^2-y^2=\operatorname{Re}z^2$, $xy=\operatorname{Re}\left(-\tfrac i2z^2\right)$, $e^x\cos y=\operatorname{Re}e^z$, $\sin x\cosh y=\operatorname{Re}\sin z$, $\frac{x}{x^2+y^2}=\operatorname{Re}\frac1z$, $\ln|z|=\operatorname{Re}\operatorname{Log}z$.
- Integrate the CR equations. $v_y=u_x$ gives $v=\int u_x\,dy+\varphi(x)$; then $v_x=-u_y$ determines $\varphi'(x)$, and $\varphi$ up to a constant.
- Milne–Thomson (Theorem 2.9): $f'(z)=u_x(z,0)-iu_y(z,0)$, then integrate in $z$. Fastest when $u$ is a polynomial or built from $e^x$, $\sin$, $\cos$.
- Given $u-v$ or $u+v$: $(1+i)f=(u-v)+i(u+v)$, so $u-v$ is the real part and $u+v$ the imaginary part of the analytic function $(1+i)f$. Find it, then divide by $1+i$.
- Fix the constant from the given value (such as $f(0)$); a real part determines $f$ only up to an added imaginary constant $iC$.
Proof. For real $x$ in the segment, Theorem 2.3 and the CR equations give $$f'(x)=u_x(x,0)+iv_x(x,0)=u_x(x,0)-iu_y(x,0).$$ So the analytic functions $f'(z)$ and $g(z)=u_x(z,0)-iu_y(z,0)$ agree on a segment of the real axis, and by the identity theorem (§2.3: two analytic functions on a domain that agree on a set with a limit point in the domain agree everywhere) they agree on $D$. The formula with $v$ follows from $f'=v_y+iv_x$ (Theorem 2.3). Integrating gives $f$ up to a constant, which must be purely imaginary since $\operatorname{Re}f=u$ is fixed. ∎
(b) Find the analytic function $f=u+iv$ with $u-v=x^3-3x^2y-3xy^2+y^3-2x-2y$ and $f(0)=0$.
- CR at a point do not give differentiability. $\sqrt{|xy|}$ and $\frac{x^3+iy^3}{x^2+y^2}$ satisfy CR at $0$ but are not differentiable there (Example 3). You also need real differentiability — e.g. continuous partials (Theorem 2.4).
- Even CR everywhere is not enough without some regularity. $f(z)=e^{-1/z^4}$, $f(0)=0$, is analytic on $\mathbb{C}\setminus\{0\}$ and satisfies CR at $0$ too (on both axes $f=e^{-1/t^4}$, so all partials at $0$ are $0$), yet it is not even continuous at $0$: on the line $z=te^{i\pi/4}$, $z^4=-t^4$ and $|f|=e^{1/t^4}\to\infty$. (The Looman–Menchoff theorem: a continuous $f$ whose partials exist and satisfy CR everywhere on $U$ is analytic on $U$.)
- Differentiable at a point $\ne$ analytic at the point. $|z|^2$ is differentiable only at $0$ and is analytic nowhere; $x^2+iy^2$ is differentiable exactly on the line $y=x$ and analytic nowhere.
- Smooth in $\mathbb{R}^2$ does not mean complex differentiable. $\bar z$, $\operatorname{Re}z$, $|z|^2$, $e^{\bar z}$ are $C^\infty$ maps of the plane; the first, second and fourth are differentiable nowhere. In particular $\bar z$ is continuous everywhere and differentiable nowhere — no Weierstrass-type construction needed.
- A real-valued function can be differentiable only where its derivative is $0$ (its derivative along the imaginary direction would be imaginary). So a nonconstant real-valued function is never analytic on a domain.
- Rigidity needs a domain. On a disconnected open set, $|f|$ or $\operatorname{Re}f$ constant gives only "constant on each component" (Example 5(c)).
- Harmonic conjugates are not symmetric. If $v$ is a conjugate of $u$, then $-u$ (not $u$) is a conjugate of $v$. $u=x^2-y^2$ has conjugate $2xy$, but $x^2-y^2$ is not a conjugate of $2xy$.
- Harmonic conjugates can fail to exist globally: $\ln|z|$ on $\mathbb{C}\setminus\{0\}$ (Theorem 2.8(3)). On a disc or any simply connected domain they always exist.
- Harmonic functions do not form a ring: $x$ is harmonic but $x^2$ is not; $u^2$ is harmonic only if $u$ is constant. (Sums and real multiples of harmonic functions are harmonic.)
- Not every function of $x,y$ that "looks complex" is analytic. Before using $f'=u_x+iv_x$ or Milne–Thomson, confirm the CR equations; e.g. $x^2-y^2+i(2xy+x)$ is not analytic.
- The polar CR equations need $r\ne0$, and $u,v$ there are functions of $(r,\theta)$ on a set where $\theta$ is a continuous function of $z$ (a cut plane, a sector).
Exercises
- 1 Evaluate or show that the limit does not exist: (a) $\displaystyle\lim_{z\to0}\frac{(\operatorname{Re}z)^2}{|z|^2}$; (b) $\displaystyle\lim_{z\to1+i}\frac{z^2-2i}{z-1-i}$; (c) $\displaystyle\lim_{z\to\infty}\frac{z^2+1}{2iz^2-z}$.
- 2 Prove from Definition 2.3 that $(z^2)'=2z$ and $\left(\frac1z\right)'=-\frac1{z^2}$ for $z\ne0$.
- 3 Show from the definition that $f(z)=\operatorname{Im}z$ is differentiable nowhere.
- 4 Show that $f(z)=\bar z^{\,2}$ is differentiable at $0$ and nowhere else.
- 5 Use the Wirtinger derivative to find where $f(z)=z|z|^2$ is differentiable, and find $f'$ there.
- 6 Show that if $f$ is differentiable at $z_0$, then $g(z)=\overline{f(\bar z)}$ is differentiable at $\bar z_0$ with $g'(\bar z_0)=\overline{f'(z_0)}$.
- 7 Prove that if a real-valued function $f$ is differentiable at $z_0$, then $f'(z_0)=0$.
- 8 Find all points where $f(z)=\left(|z|^2-1\right)^2$ is differentiable.
- 1 Verify the CR equations for $f(z)=z^3$ and for $f(z)=\cos z$, and compute $f'$ from $f'=u_x+iv_x$.
- 2 Find all points where $f=x^3+3iy$ is differentiable. Is it analytic anywhere?
- 3 Find real constants $a,b,c,d$ such that $f=x^2+axy+by^2+i(cx^2+dxy+y^2)$ is entire, and express $f$ in terms of $z$.
- 4 Use the polar CR equations to show that $f(re^{i\theta})=r^n(\cos n\theta+i\sin n\theta)$, $n\in\mathbb{Z}$, is analytic on $\mathbb{C}\setminus\{0\}$, and find $f'$.
- 5 Show that $f(z)=e^{\bar z}$ is differentiable nowhere.
- 6 Where is $f(z)=|z|^2+2z$ differentiable? Find the derivative there.
- 7 Let $f(z)=\dfrac{x^3+2y^3+i(y^3-2x^3)}{x^2+y^2}$ for $z\ne0$, $f(0)=0$. Show that $f$ satisfies the CR equations at $0$ but is not differentiable at $0$.
- 8 For $f(z)=e^{-1/z^4}$ ($z\ne0$), $f(0)=0$ (pitfall 2), show that $f$ satisfies the CR equations at every point of $\mathbb{C}$, and explain why this does not contradict Theorem 2.4.
- 9 Let $u=\frac{x}{x^2+y^2}$, $v=\frac{-y}{x^2+y^2}$ on $\mathbb{C}\setminus\{0\}$. Verify CR, compute $f'=u_x+iv_x$, and identify $f$.
- 1 Show that $u=x^3-3xy^2+3x^2-3y^2+1$ is harmonic, find a harmonic conjugate $v$, and find $f=u+iv$.
- 2 Find the analytic $f$ with $\operatorname{Re}f=\sin x\cosh y$ and $f(0)=0$.
- 3 Find a harmonic conjugate of $u=\dfrac{x}{x^2+y^2}$ on $\mathbb{C}\setminus\{0\}$.
- 4 Find the real constant $a$ for which $u=x^3+axy^2$ is harmonic, and then a harmonic conjugate.
- 5 Find the analytic $f$ with $\operatorname{Im}f=e^{-y}\sin x$ and $f(0)=1$.
- 6 Find the analytic $f=u+iv$ with $u+v=x^2-y^2+2xy+x+y$ and $f(0)=0$.
- 7 Show that if $u$ is harmonic on $U$ then $g=u_x-iu_y$ is analytic on $U$ (use only that $u$ has continuous second partials).
- 8 Show that if $u$ and $u^2$ are both harmonic on a domain, then $u$ is constant.
- 9 Describe the level curves of $u=x^2-y^2$ and $v=2xy$. Show that they are orthogonal at every point except $0$, and find the angle between the curves $u=0$ and $v=0$ at $0$.
- 1 Let $f=u+iv$ be analytic on a domain $D$ and suppose $au+bv=c$ on $D$ for real constants $a,b,c$ with $(a,b)\ne(0,0)$. Prove that $f$ is constant.
- 2 Prove the operator identity $\Delta=4\dfrac{\partial^2}{\partial z\,\partial\bar z}$ (on $C^2$ functions), and deduce that $\Delta|f|^2=4|f'|^2$ for $f$ analytic (using that $f'$ is analytic, §2.2).
- 3 Let $f$ and $g$ be analytic on a domain $D$ with $|f|^2+|g|^2$ constant. Prove that $f$ and $g$ are constant.
- 4 Let $f$ be analytic on a domain $D$ that is symmetric about the real axis. Prove that $g(z)=\overline{f(\bar z)}$ is analytic on $D$.
- 5 Let $f=u+iv$ be analytic on $\mathbb{C}$ with $u(x,y)=\varphi(x)$ depending on $x$ only. Prove that $f(z)=az+b$ with $a$ real.
- 6 Let $f$ be analytic and nowhere zero on an open set $U$. Prove that $\ln|f|$ is harmonic on $U$.
E1. CSIR Part B Let $f(x+iy)=x^3+i(1-y)^3$. Then $f$ is complex differentiable (A) nowhere; (B) only at the point $z=i$; (C) at every point of the line $\operatorname{Im}z=1$; (D) everywhere.
E2. CSIR Part B GATE MA For a real constant $k$, let $u(x,y)=2x-x^3+kxy^2$ be harmonic on $\mathbb{R}^2$. A harmonic conjugate of $u$ is (A) $2y-3x^2y+y^3$; (B) $2y+3x^2y-y^3$; (C) $-2y+3x^2y-y^3$; (D) $3x^2y-y^3$.
E3. CSIR Part B Let $f=u+iv$ be entire with $u(x,y)=e^x\cos y+x$ and $f(0)=1$. Then $f(i\pi)$ equals (A) $-1+i\pi$; (B) $1+i\pi$; (C) $-1-i\pi$; (D) $i\pi$.
E4. CSIR Part C Which of the following statements are true? (A) If $f=u+iv$ satisfies the Cauchy–Riemann equations at $z_0$, then $f$ is differentiable at $z_0$. (B) The function $f(z)=z\operatorname{Re}z$ is complex differentiable at exactly one point. (C) If $u,v$ have continuous first partial derivatives on $\mathbb{C}$ and satisfy the Cauchy–Riemann equations everywhere, then $u+iv$ is entire. (D) The function $e^x(\cos y-i\sin y)$ is entire.
E5. CSIR Part C GATE MA Which of the following are true? (A) $x^2-y^2+2xy$ is the real part of an entire function. (B) $\ln(x^2+y^2)$ is the real part of a function analytic on $\mathbb{C}\setminus\{0\}$. (C) $xy$ is harmonic on $\mathbb{R}^2$. (D) If $u$ is harmonic on $\mathbb{R}^2$, then so is $u^2$.
E6. CSIR Part C Let $f=u+iv$ be analytic on the unit disc $\mathbb{D}=\{|z|\lt1\}$. Which of the following conditions force $f$ to be constant on $\mathbb{D}$? (A) $\operatorname{Re}\big(f(z)^2\big)$ is constant on $\mathbb{D}$. (B) $\overline{f}$ is analytic on $\mathbb{D}$. (C) $u$ is constant on the segment $(-\tfrac12,\tfrac12)$ of the real axis. (D) $|f'|$ is constant on $\mathbb{D}$.
Suggested plan — 3 lectures × 60 min.
- The derivative. Definitions 2.1–2.3 and Theorem 2.1 briskly (they repeat real analysis), Example 1(a) to show that direction matters. Theorem 2.2 with the Carathéodory form — it makes the chain rule painless. Example 2 on the board: let the class guess where $|z|^2$ is differentiable before computing. Project Interactive 1 with $\bar z$ and $|z|^2$ at $z_0=1+i$, then at $z_0=0$.
- The Cauchy–Riemann equations. Theorem 2.3 with proof; the Wirtinger form as a computational shortcut. Theorem 2.4 with the full proof of (1) and the mean-value argument for (2). Example 3 — return to Interactive 1 with the function $(x^3+iy^3)/(x^2+y^2)$ at $0$: the green dot and purple ring coincide (CR hold) yet the curve refuses to shrink. Example 4, then Theorem 2.5 and the three applications.
- Rigidity and harmonic functions. Theorem 2.6 (prove (iii) carefully), Example 5. Definition 2.5, Theorem 2.7 (flag the forward reference to §2.2), Interactive 2. Theorem 2.8 with the explicit integral formula; the $\ln|z|$ counterexample. The strategy box and Milne–Thomson; Examples 6 and 7. Finish with Exam corner E4–E6 as a class quiz.
Discussion questions. (1) $\bar z$ is the simplest possible non-analytic function — it is just a reflection. Why does a reflection fail to be complex differentiable while a rotation succeeds? (2) A real function can be differentiable once but not twice; why is that impossible for a complex function on an open set, and what part of the theory is needed to prove it? (3) If you know $\operatorname{Re}f$ on a disc, how much of $f$ do you know? What if you know only $|f|$?
Misconceptions to watch for. "CR at a point ⇒ differentiable"; confusing "differentiable at $z_0$" with "analytic at $z_0$"; using $f'=u_x+iv_x$ for a function that is not analytic; writing $u$ (instead of $-u$) as the conjugate of $v$; applying Milne–Thomson to a non-harmonic $u$; forgetting that $f$ is determined by $\operatorname{Re}f$ only up to $iC$; thinking that a conjugate always exists on any domain.
Suggested homework. I(1, 4, 5, 7), II(2, 3, 6, 7), III(1, 2, 5, 6, 8), IV(1, 3, 5); Exam corner E1, E2, E6.
Research question. Harmonic functions describe steady temperatures in a metal plate, electrostatic potentials and the pressure field of groundwater. Engineers rarely have a formula — they have boundary measurements and a computer. How well does the simplest numerical method recover a harmonic function from its boundary values, and does the computed solution show the properties proved in this section?
Why it matters. Laplace's equation $\Delta u=0$ is solved millions of times a day: heat-sink design for electronics, seepage under dams, potential flow past aircraft wings, and image-processing tools that "inpaint" missing pixels by harmonic interpolation. The analytic function $f=u+iv$ is also practical: in fluid mechanics $u$ is a velocity potential and $v$ the stream function, whose level curves are the streamlines — orthogonal to the equipotentials, exactly as in Interactive 2.
What to do.
- On the unit square, impose the boundary values of $u=e^x\cos y=\operatorname{Re}e^z$. On an $n\times n$ grid ($n=11,21,41$ points per side) replace every interior value by the average of its four neighbours, repeatedly (Jacobi iteration), until the values change by less than $10^{-10}$.
- Record the number of iterations and the maximum error against the exact $u$. How does the error change when the grid spacing is halved? How does the iteration count change?
- Check the mean value property numerically: average the exact $u$ over a circle of radius $0.4$ about $(0.5,0.5)$ and compare with $u(0.5,0.5)$. (It is proved in §2.2.)
- Repeat step 1 with boundary values $x^2-y^2$. Explain why the error is now at the level of the stopping tolerance, whatever the grid.
- Compute a discrete harmonic conjugate by integrating $v_y=u_x$, $v_x=-u_y$ with finite differences along the staircase path of Theorem 2.8, and plot level curves of both.
Tools: Python with NumPy and Matplotlib (contour), or any spreadsheet for $n=11$.
Hand in: A table (grid size, iterations, maximum error), the mean-value check, a level-curve picture of $u$ and your discrete $v$, and a paragraph explaining the observed rates.
Going further. Read about successive over-relaxation (SOR), which cuts the iteration count from about $n^2$ to about $n$; about multigrid methods; and about the Kutta–Joukowski theory of lift, where the analytic function $f=u+iv$ of a flow around a wing profile gives the lift force through a contour integral (§2.2).
🧪 Self-check quiz
Q1. The function $f(z)=|z|^2$ is
Q2. If $f=u+iv$ is differentiable at $z_0$, then $f'(z_0)$ equals
Q3. A harmonic conjugate of $u=2xy$ on $\mathbb{R}^2$ is
Q4. Let $f$ be analytic on a domain $D$. Which condition does not force $f$ to be constant?
Q5. The function $u=\ln(x^2+y^2)$ on $\mathbb{C}\setminus\{0\}$ is