In this lesson you’ll extend the exponential, logarithm, trigonometric, hyperbolic and power functions to C, see which familiar properties survive and which break, and learn to handle multi-valuedness with branches and cuts.
This is forced, not chosen: it is the only definition that is analytic and agrees with ex on the real axis, by the identity theorem.
Properties that survive:
ez1+z2=ez1ez2
dzdez=ez
ez=0 for all z, since ∣ez∣=ex>0
ez is entire
The property that breaks, and it is a big one:
ez+2πi=ez
The complex exponential is periodic, with period 2πi. On R the exponential is injective; on C it is infinitely many-to-one. Every consequence in this lesson traces back to that.
Geometrically ∣ez∣=eRez and argez=Imz, so horizontal lines map to rays and vertical lines to circles. A horizontal strip of height 2π covers C∖{0} exactly once.
We want logz to invert ez. Since the exponential is not injective, there is no single answer:
logz=ln∣z∣+iargz=ln∣z∣+i(Argz+2πk),k∈Z
The complex logarithm is genuinely multi-valued, taking infinitely many values differing by 2πi. This is not a defect to be patched but a real feature, and it is the reason the subject has branch cuts, Riemann surfaces, and - much later - the residue theorem’s factor of 2πi.
The principal logarithm picks Argz∈(−π,π]:
Logz=ln∣z∣+iArgz
To make it continuous we must delete a ray from 0 to ∞. The conventional choice is the negative real axis, called the principal branch cut, and on C minus that cut, Log is analytic with
dzdLogz=z1
Two points of care that cause most of the errors in this material:
log(z1z2)=logz1+logz2 holds as sets of values, but can fail for principal values. For instance with z1=z2=−1, Log(z1z2)=Log(1)=0, while Log(−1)+Log(−1)=2πi.
log of a negative number now exists. Log(−1)=iπ, which is the honest answer to a question you were told had none.
Rearranging Euler’s formula gives the definitions:
cosz=2eiz+e−iz,sinz=2ieiz−e−iz
Both are entire, both have the expected derivatives, and every algebraic identity - sin2z+cos2z=1, the addition formulas, the double-angle formulas - carries over unchanged, because the identity theorem says an identity valid on R must persist.
The property that breaks is boundedness:
sin(iy)=isinhy,cos(iy)=coshy
both of which grow without bound. Complex sine and cosine are unbounded, and they must be, by Liouville’s theorem: a bounded entire function is constant. The familiar bound ∣sinx∣≤1 is a fact about the real axis only.
Zeros are unchanged, though: sinz=0 exactly at z=nπ, and cosz=0 exactly at z=2π+nπ, all real. Extending to C adds no new zeros.
The hyperbolic functions are the same objects rotated:
sinhz=2ez−e−z,sin(iz)=isinhz,cos(iz)=coshz
In C the distinction between trigonometric and hyperbolic functions essentially disappears, which retrospectively explains why their identities look so similar.
Because log is multi-valued, so is za in general. The number of distinct values depends on a:
a an integer: one value. Nothing new.
a=p/q in lowest terms: exactly q values, the qth roots we already know.
airrational or complex: infinitely many values.
The famous example is
ii=eilogi=ei(iπ/2+2πik)=e−π/2−2πk
Every value is real, and the principal one is e−π/2≈0.2079. That an imaginary number to an imaginary power is real is a good illustration of how much the exponential definition buys.
Also worth flagging: (za)b=zab can fail with principal branches, for the same reason as the logarithm identity. When in doubt, go back to ealogz and track the branch.
so they are multi-valued too, and need cuts. This is the cleanest explanation of why real inverse trig functions have such awkward restricted domains: those restrictions are branch choices.
Complex sine takes every complex value, which is Liouville’s theorem made concrete.
Example 6: Find all values of ii.
Solution.logi=i(2π+2πk), so
ii=eilogi=ei⋅i(π/2+2πk)=e−π/2−2πk
Every value is a positive real number, and the principal one is e−π/2≈0.2079. ∎
Example 7: Show log(z1z2)=logz1+logz2 can fail for principal values.
Solution. Take z1=z2=−1. Then z1z2=1 and
Log(1)=0,Log(−1)+Log(−1)=iπ+iπ=2πi
These differ by 2πi. ∎
The rule holds as an equality of value sets, and 2πi is a member of the set log1=2πik. The identity is true; it is the principal-value bookkeeping that fails, and that is the standard trap in this material.
AC steady state. The complex exponential eiωt replaces sines and cosines throughout electrical engineering, turning differential equations into algebra. Every phasor calculation is ez used deliberately.
Damped oscillation. A solution e(−α+iω)t combines decay and oscillation in one expression, with Re giving the decay rate and Im the frequency. Reading behaviour off the position of an exponent in C is standard in vibration and control analysis.
Quantum phases. A wavefunction’s time evolution is multiplication by e−iEt/ℏ, a pure phase of modulus 1. That the modulus is exactly 1 is conservation of probability.
Branch cuts in physics. In scattering theory and in Feynman propagators the choice of branch cut for a square root or logarithm encodes physical boundary conditions, and choosing wrongly gives an unphysical answer.
Digital signal processing. The z-transform uses complex powers z−n, and the unit circle in the z-plane is the frequency axis. Evaluating a transfer function on that circle is what a frequency response is.
Logarithmic spirals. The curve z(t)=e(a+ib)t is a logarithmic spiral, the shape of nautilus shells, spiral galaxy arms, and hurricane bands. It is the orbit of a point under repeated multiplication by a fixed complex number.