Complex Functions and Limits

What does it mean for a function of a complex variable to have a limit? A student who has mastered limits of real functions might guess the definition transfers word for word. This guess involves replacing absolute value by modulus. That guess turns out to be exactly right. But the geometric consequences are far richer than in one real dimension. Because the complex plane is two-dimensional, "approaching a point" means approaching from every direction at once. This includes spirals, curves, and any path whatsoever. This single fact makes complex analysis simultaneously more restrictive and more beautiful than real analysis. This article builds the foundational machinery: what a complex-valued function of a complex variable actually is, how to make rigorous sense of limits and continuity using the modulus as a distance, and how these ideas combine into the complex difference quotient — the object whose limit, when it exists, will later be called the derivative. Getting this machinery exactly right matters enormously. The entire edifice of complex analysis — the Cauchy–Riemann equations, contour integration, power series, the residue theorem — rests on one assumption. That assumption is that a limit taken in the plane is a limit taken from all directions simultaneously. The big picture is this: a function ff assigns to each point zz in some subset of C\mathbb{C} another complex number f(z)f(z). We can visualize ff as a map between two copies of the plane. This is sometimes drawn as two side-by-side sheets, the zz-plane and the ww-plane. Limits and continuity are defined using ε\varepsilon–δ\delta language exactly as in real analysis. The only difference is that open disks replace open intervals. The difference quotient

f(z0+h)−f(z0)h\frac{f(z_0+h)-f(z_0)}{h}

looks identical to its real counterpart. Yet demanding that its limit exist as the complex number h→0h\to 0 turns out to be an extraordinarily strong condition. By the end of this article you will be able to state precisely what it means for a complex limit to exist. You will prove the standard limit laws for complex functions. You will distinguish continuity from differentiability. You will understand why "complex differentiable" is a far more delicate notion than "real differentiable." This sets up the transition to The Cauchy-Riemann Equations and eventually to Contour Integrals.

Key takeaways
  • A complex function f:D→Cf:D\to\mathbb{C}, D⊆CD\subseteq\mathbb{C}, can be viewed as a pair of real functions u(x,y)u(x,y), v(x,y)v(x,y) via f=u+ivf = u+iv.
  • Limits use the modulus ∣z−z0∣|z-z_0| as distance; lim⁡z→z0f(z)=L\lim_{z\to z_0} f(z) = L means f(z)f(z) gets arbitrarily close to LL as zz approaches z0z_0 from any direction.
  • All the usual limit laws (sum, product, quotient, composition) hold, proved exactly as in single-variable calculus but with ∣⋅∣|\cdot| in place of absolute value.
  • Continuity at z0z_0 means lim⁡z→z0f(z)=f(z0)\lim_{z\to z_0} f(z) = f(z_0); it is equivalent to continuity of both uu and vv as functions of two real variables.
  • The complex difference quotient f(z0+h)−f(z0)h\dfrac{f(z_0+h)-f(z_0)}{h} existing as h→0h\to 0 in C\mathbb{C} (not just along the real or imaginary axis) is the seed of complex differentiability.
  • A function can be differentiable in every real-variable sense yet fail to have a complex derivative, because the complex limit must agree along infinitely many directions at once.
  • Key idea: complex limits are two-dimensional limits in disguise, and demanding a single limiting value from every direction of approach is what makes complex differentiability so much stronger than real differentiability.

Where this lives today

The definitions of complex limits, continuity, and the difference quotient are the load-bearing foundation of every subsequent tool in complex analysis. The Cauchy–Riemann equations, which characterize exactly when the difference quotient's limit exists, translate the "all-directions" requirement into two partial differential equations linking uu and vv — this is the immediate next step built on this article. Contour integrals, and ultimately Cauchy's integral theorem and formula, depend on functions being continuous (indeed holomorphic) along paths in the plane, using precisely the continuity notion defined here. Beyond pure mathematics, complex limits and continuity underpin: fluid dynamics and aerodynamics, where holomorphic velocity potentials model idealized 2D flow around airfoils; electrical engineering, where impedance and transfer functions are complex-valued functions of a complex frequency variable ss, and stability analysis hinges on continuity and limiting behavior as ss approaches the imaginary axis; control theory, where Nyquist and root-locus techniques trace continuous images of contours under rational transfer functions; signal processing, where the zz-transform and discrete Fourier analysis rely on convergence of complex power series; quantum mechanics and statistical physics, where analytic continuation of correlation functions and partition functions across the complex plane requires exactly this limit machinery; and cryptography and number theory, where the Riemann zeta function's analytic continuation and the study of its zeros depend fundamentally on complex limits and continuity established at this foundational level.

Ask the tutor: If u(x,y)u(x,y) and v(x,y)v(x,y) are both continuous everywhere, must f=u+ivf=u+iv be continuous as a complex function — and why doesn't the same logic force ff to be complex differentiable?

DefinitionComplex function, limit, continuity, and difference quotient

Let D⊆CD \subseteq \mathbb{C} be a nonempty set. A complex function on DD is a rule

f:D→C,z↦f(z),f: D \to \mathbb{C}, \qquad z \mapsto f(z),

assigning to each z∈Dz \in D a unique complex number f(z)f(z). Writing z=x+iyz = x+iy and f(z)=u(x,y)+iv(x,y)f(z) = u(x,y) + iv(x,y) with u,v:D→Ru,v:D\to\mathbb{R}, we call uu the real part and vv the imaginary part of ff. A point z0∈Cz_0 \in \mathbb{C} is a limit point (accumulation point) of DD if every open disk centered at z0z_0 contains a point of DD other than z0z_0. Let z0z_0 be a limit point of DD and let L∈CL \in \mathbb{C}. We say

lim⁡z→z0f(z)=L\lim_{z \to z_0} f(z) = L

if for every ε>0\varepsilon > 0 there exists δ>0\delta > 0 such that

0<∣z−z0∣<δ  and  z∈D⟹∣f(z)−L∣<ε.0 < |z - z_0| < \delta \ \text{ and } \ z \in D \quad \Longrightarrow \quad |f(z) - L| < \varepsilon.

If z0∈Dz_0 \in D, we say ff is continuous at z0z_0 if

lim⁡z→z0f(z)=f(z0).\lim_{z\to z_0} f(z) = f(z_0).

ff is continuous on DD if it is continuous at every point of DD. Finally, suppose DD is open and z0∈Dz_0 \in D. The complex difference quotient of ff at z0z_0 is the function of hh (for h≠0h \ne 0, z0+h∈Dz_0+h \in D)

Q(h)=f(z0+h)−f(z0)h.Q(h) = \frac{f(z_0+h) - f(z_0)}{h}.

If lim⁡h→0Q(h)\lim_{h\to 0} Q(h) exists (as a limit of a complex function of the complex variable hh, in the sense above), we call this limit the derivative f′(z0)f'(z_0), and say ff is complex differentiable at z0z_0.

Intuition

The crucial word in the definition above is disk. In single-variable real calculus, "x→x0x \to x_0" means xx approaches x0x_0 along the one-dimensional real line — there are only two directions, left and right, and a limit exists precisely when both one-sided limits agree. In the complex plane, z0z_0 has a whole disk of neighbors, and "z→z0z \to z_0" allows approach along straight lines at any angle, along spirals, along wiggly curves, along sequences that jump around erratically — literally any way of choosing points zn∈Dz_n \in D with zn→z0z_n \to z_0. The ε\varepsilon–δ\delta definition forces f(z)f(z) to converge to the same value LL regardless of which of these infinitely many approach patterns we use. This is why checking a complex limit by restricting to the real axis or to the imaginary axis is not enough — famously, verifying limits along two paths can agree while the full two-dimensional limit fails to exist, exactly as with the multivariable real limits studied in vector calculus. Below, the first picture shows the "approach from all directions" idea for a limit at a point z0z_0; the second shows how the difference quotient's defining ratio depends on the direction of the increment hh, which is the heart of why complex differentiability is so restrictive.

Figure
A punctured disk of radius δ\delta around z0z_0: every path of approach must land inside the ε\varepsilon-disk around LL
z-plane z₀ |z−z₀| < δ → f w-plane L |f(z)−L| < ε
Figure
The difference quotient measures rotation and stretching along a direction θ\theta; complex differentiability requires the same limiting value for every θ\theta
Increments h = re^{iθ} at z₀ z₀ θ=0 θ=π/2 θ=3π/4 θ=-π/4 Q(h) = [f(z₀+h) − f(z₀)] / h must → the same L for every θ

Central results

TheoremAlgebra of complex limits

Let D⊆CD \subseteq \mathbb{C}, let z0z_0 be a limit point of DD, let f,g:D→Cf,g:D\to\mathbb{C}, and suppose

lim⁡z→z0f(z)=L,lim⁡z→z0g(z)=M\lim_{z\to z_0} f(z) = L, \qquad \lim_{z\to z_0} g(z) = M

for complex numbers L,ML, M. Then: 1. lim⁡z→z0(f(z)+g(z))=L+M\displaystyle \lim_{z\to z_0} \big(f(z) + g(z)\big) = L + M. 2. lim⁡z→z0(f(z) g(z))=LM\displaystyle \lim_{z\to z_0} \big(f(z)\, g(z)\big) = LM. 3. If M≠0M \ne 0, then lim⁡z→z0f(z)g(z)=LM\displaystyle \lim_{z\to z_0} \frac{f(z)}{g(z)} = \frac{L}{M} (defined on the set where g(z)≠0g(z)\ne 0, which still has z0z_0 as a limit point since g(z)→M≠0g(z)\to M\ne0 forces g≠0g\ne0 near z0z_0). 4. ff is continuous at z0∈Dz_0 \in D if and only if u=Re⁡fu=\operatorname{Re} f and v=Im⁡fv = \operatorname{Im} f are both continuous at z0z_0 as functions of the two real variables x,yx,y.

Proof

We use the ε\varepsilon–δ\delta method, identical in structure to the real-variable proofs but with ∣⋅∣|\cdot| denoting the complex modulus, together with the triangle inequality ∣a+b∣≤∣a∣+∣b∣|a+b|\le |a|+|b| and its consequence ∣ab∣=∣a∣∣b∣|ab| = |a||b|.

(1) Sum rule. Let ε>0\varepsilon>0. Since f(z)→Lf(z)\to L, choose δ1>0\delta_1>0 so that 0<∣z−z0∣<δ10<|z-z_0|<\delta_1, z∈Dz\in D implies ∣f(z)−L∣<ε/2|f(z)-L|<\varepsilon/2. Since g(z)→Mg(z)\to M, choose δ2>0\delta_2>0 so that 0<∣z−z0∣<δ20<|z-z_0|<\delta_2, z∈Dz\in D implies ∣g(z)−M∣<ε/2|g(z)-M|<\varepsilon/2. Let δ=min⁡(δ1,δ2)>0\delta = \min(\delta_1,\delta_2)>0. If 0<∣z−z0∣<δ0<|z-z_0|<\delta and z∈Dz\in D, then both inequalities hold, and by the triangle inequality

∣(f(z)+g(z))−(L+M)∣=∣(f(z)−L)+(g(z)−M)∣≤∣f(z)−L∣+∣g(z)−M∣<ε2+ε2=ε.|(f(z)+g(z)) - (L+M)| = |(f(z)-L)+(g(z)-M)| \le |f(z)-L| + |g(z)-M| < \frac{\varepsilon}{2}+\frac{\varepsilon}{2} = \varepsilon.

This proves the sum rule.

(2) Product rule. First note that convergence implies local boundedness: since f(z)→Lf(z)\to L, taking ε=1\varepsilon = 1 gives some δ0>0\delta_0>0 with ∣f(z)−L∣<1|f(z)-L|<1 for 0<∣z−z0∣<δ00<|z-z_0|<\delta_0, z∈Dz\in D; hence by the triangle inequality ∣f(z)∣≤∣f(z)−L∣+∣L∣<1+∣L∣|f(z)| \le |f(z)-L|+|L| < 1+|L| on this punctured disk. Set C=1+∣L∣C = 1+|L|. Now let ε>0\varepsilon>0. Choose δ1≤δ0\delta_1 \le \delta_0 so that 0<∣z−z0∣<δ10<|z-z_0|<\delta_1, z∈Dz\in D gives ∣f(z)−L∣<ε2(∣M∣+1)|f(z)-L| < \dfrac{\varepsilon}{2(|M|+1)}, and choose δ2>0\delta_2>0 so that 0<∣z−z0∣<δ20<|z-z_0|<\delta_2, z∈Dz\in D gives ∣g(z)−M∣<ε2C|g(z)-M| < \dfrac{\varepsilon}{2C}. Let δ=min⁡(δ1,δ2)\delta=\min(\delta_1,\delta_2). For 0<∣z−z0∣<δ0<|z-z_0|<\delta, z∈Dz \in D, write the standard algebraic identity

f(z)g(z)−LM=f(z)(g(z)−M)+M(f(z)−L).f(z)g(z) - LM = f(z)\big(g(z)-M\big) + M\big(f(z)-L\big).

Taking moduli and using the triangle inequality and ∣f(z)∣<C|f(z)|<C:

∣f(z)g(z)−LM∣≤∣f(z)∣ ∣g(z)−M∣+∣M∣ ∣f(z)−L∣<C⋅ε2C+∣M∣⋅ε2(∣M∣+1)<ε2+ε2=ε.|f(z)g(z)-LM| \le |f(z)|\,|g(z)-M| + |M|\,|f(z)-L| < C\cdot\frac{\varepsilon}{2C} + |M|\cdot\frac{\varepsilon}{2(|M|+1)} < \frac{\varepsilon}{2}+\frac{\varepsilon}{2}=\varepsilon.

This proves the product rule.

(3) Quotient rule. It suffices (by the product rule, applied to f(z)f(z) and 1/g(z)1/g(z)) to show 1/g(z)→1/M1/g(z) \to 1/M when M≠0M\ne 0. Since g(z)→M≠0g(z)\to M \ne 0, take ε1=∣M∣/2>0\varepsilon_1 = |M|/2 > 0: there is δ1>0\delta_1>0 with 0<∣z−z0∣<δ10<|z-z_0|<\delta_1, z∈Dz\in D implying ∣g(z)−M∣<∣M∣/2|g(z)-M|<|M|/2, hence by the reverse triangle inequality ∣g(z)∣≥∣M∣−∣g(z)−M∣>∣M∣/2>0|g(z)| \ge |M|-|g(z)-M| > |M|/2 > 0. This shows g(z)≠0g(z)\ne 0 on this punctured disk, so 1/g(z)1/g(z) is defined there, and z0z_0 remains a limit point of D∩{g≠0}D\cap\{g\ne0\}. Now let ε>0\varepsilon>0. Choose δ2≤δ1\delta_2 \le \delta_1 so that 0<∣z−z0∣<δ20<|z-z_0|<\delta_2, z∈Dz\in D implies ∣g(z)−M∣<ε∣M∣22|g(z)-M| < \dfrac{\varepsilon |M|^2}{2}. For such zz,

∣1g(z)−1M∣=∣M−g(z)∣∣g(z)∣∣M∣<ε∣M∣2/2(∣M∣/2)∣M∣=ε.\left|\frac{1}{g(z)} - \frac{1}{M}\right| = \frac{|M-g(z)|}{|g(z)||M|} < \frac{\varepsilon|M|^2/2}{(|M|/2)|M|} = \varepsilon.

This proves 1/g(z)→1/M1/g(z)\to 1/M, and combining with the product rule gives f(z)/g(z)→L/Mf(z)/g(z) \to L/M.

(4) Continuity and real/imaginary parts. Write f=u+ivf = u+iv, L=a+ibL = a+ib where L=f(z0)L = f(z_0) if we are proving continuity, i.e. a=u(x0,y0)a=u(x_0,y_0), b=v(x0,y0)b=v(x_0,y_0). Two elementary modulus inequalities hold for any complex number w=p+iqw=p+iq:

∣p∣≤∣w∣,∣q∣≤∣w∣,∣w∣≤∣p∣+∣q∣.|p| \le |w|, \qquad |q|\le |w|, \qquad |w| \le |p|+|q|.

The first two follow since ∣w∣2=p2+q2≥p2|w|^2 = p^2+q^2 \ge p^2 and ≥q2\ge q^2. The third is the triangle inequality applied to w=p+iqw = p + iq (viewing pp and iqiq as two complex numbers summing to ww). (⇒\Rightarrow) Suppose f(z)→Lf(z)\to L as z→z0z\to z_0. Given ε>0\varepsilon>0, choose δ>0\delta>0 as in the definition of the limit of ff. Then for (x,y)(x,y) with 0<(x−x0)2+(y−y0)2<δ0<\sqrt{(x-x_0)^2+(y-y_0)^2}<\delta, writing z=x+iyz=x+iy, we have ∣f(z)−L∣<ε|f(z)-L|<\varepsilon, hence by the first inequality above (applied to w=f(z)−L=(u−a)+i(v−b)w = f(z)-L = (u-a)+i(v-b)),

∣u(x,y)−a∣≤∣f(z)−L∣<ε,∣v(x,y)−b∣≤∣f(z)−L∣<ε.|u(x,y)-a| \le |f(z)-L| < \varepsilon, \qquad |v(x,y)-b|\le |f(z)-L|<\varepsilon.

So u→au\to a and v→bv \to b as (x,y)→(x0,y0)(x,y)\to(x_0,y_0); that is, u,vu,v are continuous at (x0,y0)(x_0,y_0) (once we also check z0∈Dz_0\in D gives u(x0,y0)=au(x_0,y_0)=a, v(x0,y0)=bv(x_0,y_0)=b trivially from f(z0)=Lf(z_0)=L). (⇐\Leftarrow) Suppose u→au\to a and v→bv\to b as (x,y)→(x0,y0)(x,y)\to(x_0,y_0). Given ε>0\varepsilon>0, choose δ1>0\delta_1>0 with ∣u(x,y)−a∣<ε/2|u(x,y)-a|<\varepsilon/2 and δ2>0\delta_2>0 with ∣v(x,y)−b∣<ε/2|v(x,y)-b|<\varepsilon/2 whenever 0<(x−x0)2+(y−y0)2<δi0<\sqrt{(x-x_0)^2+(y-y_0)^2}<\delta_i. Let δ=min⁡(δ1,δ2)\delta=\min(\delta_1,\delta_2). If 0<∣z−z0∣<δ0<|z-z_0|<\delta then 0<(x−x0)2+(y−y0)2<δ0<\sqrt{(x-x_0)^2+(y-y_0)^2}<\delta (since ∣z−z0∣|z-z_0| is exactly that Euclidean distance), so by the third inequality above,

∣f(z)−L∣=∣(u−a)+i(v−b)∣≤∣u−a∣+∣v−b∣<ε2+ε2=ε.|f(z)-L| = |(u-a)+i(v-b)| \le |u-a|+|v-b| < \frac{\varepsilon}{2}+\frac{\varepsilon}{2}=\varepsilon.

Hence f(z)→Lf(z)\to L. This completes the proof of (4), and of the theorem. ■\blacksquare

∎
CorollaryContinuity is preserved by algebraic operations and composition

If f,gf,g are continuous at z0z_0, then so are f+gf+g, fgfg, and (if g(z0)≠0g(z_0)\ne0) f/gf/g. If ff is continuous at z0z_0 and gg is continuous at f(z0)f(z_0), then g∘fg\circ f is continuous at z0z_0. Consequently every polynomial p(z)=anzn+⋯+a0p(z) = a_n z^n + \cdots + a_0 is continuous on all of C\mathbb{C}, and every rational function is continuous wherever its denominator is nonzero.

Proof

The algebraic closure statements follow immediately from parts (1)–(3) of the theorem by taking L=f(z0)L=f(z_0), M=g(z0)M=g(z_0) (continuity at z0z_0 is the statement that the limit as z→z0z\to z_0 equals the value at z0z_0). For composition: let ε>0\varepsilon>0. Since gg is continuous at w0:=f(z0)w_0:=f(z_0), there is η>0\eta>0 such that ∣w−w0∣<η⇒∣g(w)−g(w0)∣<ε|w-w_0|<\eta \Rightarrow |g(w)-g(w_0)|<\varepsilon (allowing w=w0w=w_0 trivially). Since ff is continuous at z0z_0, there is δ>0\delta>0 such that ∣z−z0∣<δ⇒∣f(z)−w0∣<η|z-z_0|<\delta \Rightarrow |f(z)-w_0|<\eta. Combining, ∣z−z0∣<δ⇒∣g(f(z))−g(w0)∣<ε|z-z_0|<\delta \Rightarrow |g(f(z)) - g(w_0)| < \varepsilon, i.e. g∘fg\circ f is continuous at z0z_0. For polynomials: the identity function z↦zz\mapsto z is continuous everywhere (given ε>0\varepsilon>0 take δ=ε\delta=\varepsilon; then ∣z−z0∣<δ|z-z_0|<\delta trivially gives ∣z−z0∣<ε|z-z_0|<\varepsilon). Constant functions are continuous everywhere (given any ε>0\varepsilon>0, any δ\delta works since ∣c−c∣=0<ε|c-c|=0<\varepsilon). Repeated application of the product rule shows z↦znz\mapsto z^n is continuous for every n≥1n\ge 1, and repeated application of the sum rule together with continuity of constant multiples (a special case of the product rule with g≡cg\equiv c) shows any finite linear combination anzn+⋯+a0a_n z^n+\cdots+a_0 is continuous everywhere. Rational functions are quotients of polynomials, so the quotient rule finishes the claim wherever the denominator is nonzero. ■\blacksquare

∎

Worked examples

Example 1: A limit computed directly by algebraic manipulation. Compute

lim⁡z→2−iz2−(2−i)2z−(2−i).\lim_{z \to 2-i} \frac{z^2 - (2-i)^2}{z-(2-i)}.

Let z0=2−iz_0 = 2-i. For z≠z0z \ne z_0 the standard factorization z2−z02=(z−z0)(z+z0)z^2 - z_0^2 = (z-z_0)(z+z_0) holds for complex numbers exactly as for real ones (it is a purely algebraic identity valid in any commutative ring), so

z2−z02z−z0=z+z0(z≠z0).\frac{z^2-z_0^2}{z-z_0} = z + z_0 \qquad (z \ne z_0).

The function z↦z+z0z\mapsto z+z_0 is continuous everywhere (sum of the identity function and a constant, by the Corollary), so

lim⁡z→z0(z+z0)=z0+z0=2z0=2(2−i)=4−2i.\lim_{z\to z_0}(z+z_0) = z_0 + z_0 = 2z_0 = 2(2-i) = 4-2i.

Since the original quotient agrees with z+z0z+z_0 for all z≠z0z\ne z_0, and limits only depend on the function's values near but not at z0z_0, the original limit equals 4−2i4-2i as well. As a check, let us also verify directly: with z=x+iyz = x+iy, z0=2−iz_0=2-i, we can expand z2−(2−i)2=z2−3+4iz^2-(2-i)^2 = z^2-3+4i and confirm numerically at a nearby test point z=1.99−1.01iz=1.99-1.01i:

z2=(1.99)2−(1.01)2+2(1.99)(−1.01)i=3.9601−1.0201−4.0198i=2.9400−4.0198i,z^2 = (1.99)^2-(1.01)^2 + 2(1.99)(-1.01)i = 3.9601-1.0201 -4.0198i = 2.9400 - 4.0198i,
z2−3+4i=−0.0600−0.0198i,z−z0=−0.01−0.01i,z^2-3+4i = -0.0600-0.0198i, \qquad z - z_0 = -0.01 - 0.01i,
z2−3+4iz−z0=−0.0600−0.0198i−0.01−0.01i.\frac{z^2-3+4i}{z-z_0} = \frac{-0.0600-0.0198i}{-0.01-0.01i}.

Multiplying numerator and denominator by the conjugate −0.01+0.01i-0.01+0.01i of the denominator: denominator becomes (−0.01)2+(0.01)2=0.0002(-0.01)^2+(0.01)^2 = 0.0002, and numerator becomes

(−0.0600−0.0198i)(−0.01+0.01i)=0.0006−0.0006i+0.000198i−0.000198i2(-0.0600-0.0198i)(-0.01+0.01i) = 0.0006 - 0.0006i + 0.000198i - 0.000198 i^2
=0.0006+0.000198+(−0.0006+0.000198)i=0.000798−0.000402i.= 0.0006 + 0.000198 + (-0.0006+0.000198)i = 0.000798 - 0.000402i.

Dividing by 0.00020.0002:  3.99−2.01i\ 3.99 - 2.01i, indeed very close to 4−2i4-2i, confirming the algebraic computation.

Example 2: A limit that fails to exist because different directions of approach give different values (and the corresponding difference-quotient failure). Consider f(z)=zˉzf(z) = \dfrac{\bar z}{z} for z≠0z\ne 0 (recall zˉ=x−iy\bar z = x-iy is the complex conjugate of z=x+iyz=x+iy). Does lim⁡z→0f(z)\displaystyle\lim_{z\to 0} f(z) exist? Approach along the positive real axis: let z=tz = t with t>0t>0, t→0+t\to 0^+. Then zˉ=t\bar z = t, so f(z)=t/t=1f(z) = t/t = 1 for all such tt, giving limiting value 11. Approach along the positive imaginary axis: let z=itz = it with t>0t>0, t→0+t \to 0^+. Then zˉ=−it\bar z = -it, so

f(z)=−itit=−1f(z) = \frac{-it}{it} = -1

for all such tt, giving limiting value −1-1. Since these two paths of approach to z0=0z_0=0 produce different values (1≠−11 \ne -1), the definition of the limit — which requires one value LL working for every sequence of points approaching 00 inside a punctured disk of radius δ\delta — cannot be satisfied for any candidate LL: whichever LL we propose, at least one of these two paths eventually leaves any ε\varepsilon-neighborhood of LL for ε<1\varepsilon < 1. (Concretely: if L=1L=1 take ε=1/2\varepsilon = 1/2; along the imaginary-axis path every value equals −1-1, and ∣−1−1∣=2>1/2|-1-1| = 2 > 1/2, so no δ\delta works. A symmetric argument rules out every other candidate LL.) Hence

lim⁡z→0zˉzdoes not exist.\lim_{z\to 0} \frac{\bar z}{z} \quad \text{does not exist.}

This example is not merely a curiosity: it is exactly the difference quotient of g(z)=zˉg(z) = \bar z at z0=0z_0 = 0, since

Q(h)=g(0+h)−g(0)h=hˉ−0h=hˉh.Q(h) = \frac{g(0+h)-g(0)}{h} = \frac{\bar h - 0}{h} = \frac{\bar h}{h}.

We have just shown lim⁡h→0Q(h)\lim_{h\to 0} Q(h) does not exist, so g(z)=zˉg(z)=\bar z is not complex differentiable at z=0z=0 — despite u(x,y)=xu(x,y)=x and v(x,y)=−yv(x,y)=-y being perfectly smooth, differentiable real functions of two variables. This is the first hint of how much stronger complex differentiability is than ordinary real-variable smoothness; the precise mechanism behind this failure is explained by The Cauchy-Riemann Equations.

CounterexampleWhy "limit exists along two paths and they agree" does not imply the limit exists

One might be tempted to weaken the definition of complex limit to: "f(z)→Lf(z)\to L if f(z)→Lf(z)\to L along the real axis and along the imaginary axis." This is not equivalent to the correct ε\varepsilon–δ\delta definition, and dropping the "for every path of approach" requirement breaks the theorem's proof machinery (in particular, the product and quotient rules used repeatedly above rely on the full disk-based definition). Consider

f(z)=xyx2+y2(z=x+iy≠0),f(z) = \frac{xy}{x^2+y^2} \qquad (z = x+iy \ne 0),

a real-valued (hence complex-valued) function. Along the real axis (y=0y=0, x→0x\to 0): f(z)=0f(z) = 0 for all x≠0x\ne 0, so the limit along this path is 00. Along the imaginary axis (x=0x=0, y→0y\to 0): f(z)=0f(z) = 0 for all y≠0y \ne 0, so the limit along this path is also 00. The two axis limits agree, both equal to 00. Yet the true two-dimensional limit does not exist. Approach along the line y=xy=x (i.e. z=t(1+i)z = t(1+i), t→0t\to 0, t≠0t \ne 0):

f(z)=t⋅tt2+t2=t22t2=12f(z) = \frac{t\cdot t}{t^2+t^2} = \frac{t^2}{2t^2} = \frac{1}{2}

for every such tt, so the limit along this diagonal path is 12≠0\tfrac12 \ne 0. Since two different paths of approach to 00 give different limiting values (00 and 12\tfrac12), the genuine limit lim⁡z→0f(z)\lim_{z\to 0} f(z) fails to exist by the same reasoning as in Example 2. This shows precisely why the correct definition demands the ε\varepsilon–δ\delta condition hold for every point in a punctured disk simultaneously, not merely along finitely many test paths: checking only the real and imaginary axes (dropping the "all directions" requirement) is a strictly weaker condition that can be satisfied even when the true complex limit does not exist. Any theorem "proved" using only axis-checks — for instance, a false claim that ff above is continuous at 00 once f(0)f(0) is defined to be 00 — would be invalid, because continuity there fails: along y=xy=x the function tends to 1/2≠f(0)1/2 \ne f(0).

Scope, edge cases & common traps

SituationWhat is trueWhat commonly goes wrong
z0z_0 not a limit point of DDThe limit definition is vacuous or ill-posed; one should only speak of limits at accumulation points of the domainStudents write limits at isolated points of a discrete domain, where the condition is trivially/vacuously satisfiable for any LL
Checking a limit along pathsAgreement along finitely many paths (e.g. real & imaginary axes) never proves a limit existsConcluding existence from two matching path-limits, as shown in the counterexample above
Continuity vs. differentiabilityContinuity of u,vu,v (part 4 of the theorem) is much weaker than complex differentiabilityAssuming smooth real and imaginary parts automatically give a complex derivative (zˉ\bar z is the standard cautionary example)
Quotient rule domainz0z_0 must remain a limit point of {z∈D:g(z)≠0}\{z \in D : g(z)\ne 0\}Forgetting to verify g≠0g\ne 0 near z0z_0 before invoking the quotient rule
Difference quotient directionh→0h\to 0 must be allowed to approach 00 from every complex directionOnly testing hh real or hh purely imaginary and declaring the derivative exists

The mathematician behind it

Origins · 1 minAugustin-Louis Cauchy and how this was discovered
AC
Augustin-Louis Cauchy
1789–1857 · French mathematician, École Polytechnique and Collège de France

Cauchy was the principal architect of rigorous ε\varepsilon–δ\delta-style analysis and, alongside Riemann and Weierstrass, one of the founders of complex function theory. Working amid the turbulence of post-Revolutionary and post-Napoleonic France — he was at one point exiled for his royalist sympathies — Cauchy produced an enormous body of work: over 800 papers, spanning analysis, mathematical physics, elasticity theory, and the foundations of calculus. His insistence on precise limit definitions, rather than the looser "infinitesimal" reasoning of the eighteenth century, is exactly the discipline embodied in the modern definitions of complex limits and continuity presented here. Cauchy's investigations into what would become the Cauchy–Riemann equations and the Cauchy integral theorem grew directly out of applying his rigorous real-variable limit concept to functions of a complex variable, treating C\mathbb{C} as a genuine metric space long before that language existed.

History
- 1814: Cauchy presents early work on definite integrals with complex limits of integration, planting seeds of complex analysis.
- 1821: Cauchy's Cours d'Analyse rigorously formalizes limits, continuity, and convergence for real functions, providing the template later transplanted to C\mathbb{C}.
- 1825: Cauchy's memoir on integrals between imaginary limits lays groundwork for what becomes contour integration.
- 1851: Riemann's doctoral dissertation formulates the Cauchy–Riemann equations and geometric function theory, building on Cauchy's limit and continuity framework.
- 1874: Weierstrass's lectures popularize power-series-based, fully rigorous treatments of complex differentiability, cementing the ε\varepsilon–δ\delta approach used today.
Is a complex limit just two real limits happening at once?

Yes, in a precise sense: Theorem part (4) shows lim⁡z→z0f(z)=L\lim_{z\to z_0} f(z) = L (with f=u+ivf=u+iv, L=a+ibL=a+ib) if and only if u(x,y)→au(x,y)\to a and v(x,y)→bv(x,y)\to b as (x,y)→(x0,y0)(x,y)\to(x_0,y_0), using the ordinary two-variable real limit. So computing a complex limit reduces to computing two simultaneous real, two-variable limits. However, this equivalence is about existence of the limit as a package, not a shortcut for definitions like differentiability: the complex derivative additionally requires the difference quotient (a specific combination of u,vu,v and their partials) to converge to one complex number regardless of direction, which is a strictly stronger condition than mere continuity of u,vu,v, as the zˉ\bar z example shows.

Why can't I just check limits along the real and imaginary axes and call it done?

Because a genuine complex limit requires convergence to the same value along every possible path of approach in the plane, not just two special directions. The counterexample with f(z)=xy/(x2+y2)f(z)=xy/(x^2+y^2) shows two paths (the axes) can agree perfectly while a third path (the diagonal y=xy=x) gives an entirely different value, so the true limit does not exist. Checking finitely many paths can only disprove a limit (by finding two paths that disagree); it can never prove one exists, since there are infinitely many paths of approach in two dimensions and no finite check can rule out disagreement along some path you didn't test.

How is the complex difference quotient different from the real one?

Algebraically the formula (f(z0+h)−f(z0))/h\big(f(z_0+h)-f(z_0)\big)/h looks identical to the real-variable difference quotient. The difference is that hh is now a complex number, so h→0h\to 0 means hh can approach 00 along any direction in the plane — real, imaginary, or any angle — and even along curving paths. Demanding the quotient converge to one value f′(z0)f'(z_0) under all these approaches is a much stronger requirement than the real case, where h→0h\to0 only has two directions. This extra strength is what eventually forces uu and vv to satisfy the Cauchy–Riemann equations whenever ff is complex differentiable.

Does continuity of ff imply the complex derivative exists?

No. Continuity only requires f(z)→f(z0)f(z)\to f(z_0) as z→z0z\to z_0, a comparatively mild condition equivalent to continuity of uu and vv. Complex differentiability requires the much sharper condition that the difference quotient itself converges, which involves a subtraction and division by hh that can amplify small differences in direction-dependent behavior. The function f(z)=zˉf(z)=\bar z is continuous everywhere (since u=xu=x, v=−yv=-y are continuous), yet its difference quotient at every point fails to converge, exactly as shown in Example 2, so it is nowhere complex differentiable.

What role does the modulus ∣z−z0∣|z-z_0| play compared to absolute value in real analysis?

The modulus ∣z−z0∣=(x−x0)2+(y−y0)2|z-z_0| = \sqrt{(x-x_0)^2+(y-y_0)^2} is simply the Euclidean distance between the two points of the plane representing zz and z0z_0. It plays exactly the role that ∣x−x0∣|x-x_0| plays on the real line: it measures "closeness" and appears in the ε\varepsilon–δ\delta conditions defining limits and continuity. The key structural fact making the proofs of the limit laws go through unchanged is that ∣⋅∣|\cdot| on C\mathbb{C} satisfies the same triangle inequality and multiplicativity (∣zw∣=∣z∣∣w∣|zw|=|z||w|) that ∣⋅∣|\cdot| satisfies on R\mathbb{R}, so every ε\varepsilon–δ\delta argument from single-variable calculus transfers essentially verbatim.

Why is this called "Tier 1" material if it already anticipates deep results like Cauchy–Riemann?

This article only develops the language and elementary machinery — definitions of function, limit, continuity, and difference quotient, plus their basic algebraic properties — using nothing beyond properties of the complex modulus already established when Complex Numbers and the Complex Plane were introduced. It does not derive the Cauchy–Riemann equations, prove any integral theorems, or develop power series; it only sets up the precise vocabulary those later, more advanced results are stated and proved in. The depth of the eventual consequences does not make the foundational definitions themselves advanced.

Can z0z_0 or LL be "at infinity" in these definitions?

Not in the definitions given here, which are stated for finite z0,L∈Cz_0, L \in \mathbb{C} using the Euclidean modulus. A rigorous treatment of limits involving the point at infinity requires introducing the extended complex plane C^=C∪{∞}\widehat{\mathbb{C}} = \mathbb{C}\cup\{\infty\} and the chordal (or spherical) metric associated to the Riemann sphere, which is a distinct, more advanced construction. The finite-point theory developed above is the necessary prerequisite for that extension.

Before this

Where this leads