Natural Logarithm: ln(x)

What ln means, its derivative, and why it's used in AI loss functions.

What is ln?

ln(x) = natural logarithm = log base e

ln answers: "e to what power gives x?"

ln(1)    = 0     because e⁰ = 1
ln(e)    = 1     because e¹ = e
ln(7.39) = 2     because e² ≈ 7.389
ln(0.5)  = -0.69 because e^(-0.69) ≈ 0.5

Derivative of ln

f(x) = ln(x)
f'(x) = 1/x

At x=1: f'(1) = 1/1 = 1
At x=2: f'(2) = 1/2 = 0.5
At x=5: f'(5) = 1/5 = 0.2
At x=0.1: f'(0.1) = 1/0.1 = 10  ← steep near zero!

Why ln Appears in AI Loss Functions

Cross-Entropy Loss = -y·ln(ŷ)

If target y=1:
  ŷ=0.9 (good):  Loss = -ln(0.9) = 0.105   (small ✓)
  ŷ=0.5 (okay):  Loss = -ln(0.5) = 0.693   (medium)
  ŷ=0.1 (bad):   Loss = -ln(0.1) = 2.303   (big! ✗)
  ŷ=0.001 (terrible): Loss = -ln(0.001) = 6.908 (huge!! ✗✗)

ln punishes bad predictions HARD.
Gentle for good predictions.

Why ln Instead of Squared Error?

Loss graph (-ln(ŷ) when target=1):

  Loss ▲
   7   │·                        ← terrible prediction: huge penalty
   5   │ ·
   3   │   ··
   2   │     ···
   1   │        ····
   0   │            ·········    ← good prediction: almost no penalty
       └──────────────────────► ŷ
       0   0.2  0.4  0.6  0.8  1.0

Gradient = -1/ŷ:
  ŷ=0.9:  gradient = -1.11   (small push)
  ŷ=0.1:  gradient = -10     (big push!)
  ŷ=0.01: gradient = -100    (huge push!!)

→ Worse prediction = stronger correction
→ Model fixes big mistakes FAST

Limits: lim(x)

"What value does the function approach?"

What lim Means

lim f(x) = L
x→a

"As x approaches a, f(x) approaches L"

It's the DESTINATION, even if you never arrive.

Simple Example

lim (x + 1) = ?
x→3

  x=2.9:   3.9
  x=2.99:  3.99
  x=2.999: 3.999
  x→3:     → 4

Answer: 4 (just plug in)

When You CAN'T Plug In (0/0)

lim (x² - 1)/(x - 1) = ?
x→1

Plug in: (1-1)/(1-1) = 0/0 ← undefined!

But approach from both sides:
  x=0.99:  → 1.99
  x=0.999: → 1.999
  x=1.001: → 2.001
  x=1.01:  → 2.01

Approaches 2!

Why: (x²-1)/(x-1) = (x+1)(x-1)/(x-1) = x+1
At x→1: x+1 → 2 ✓

Limit with Infinity

lim 1/x = ?
x→∞

  x=10:      0.1
  x=100:     0.01
  x=1000:    0.001
  x=1000000: 0.000001

Approaches 0. Never reaches it.
lim 1/x = 0
x→∞

Euler's Number AS a Limit

e = lim (1 + 1/n)ⁿ
    n→∞

  n=1:      2.000
  n=10:     2.594
  n=100:    2.705
  n=1000:   2.717
  n=10000:  2.7181
  n→∞:      → 2.71828... = e

Approaches e but never passes it.

  ▲ value
e │─ ─ ─ ─ ─ ─ ─ ─ ─ ← limit line
  │          ········
  │       ··
  │     ·
  │   ·
  │ ·
2 │·
  └──────────────────► n
  1   10  100 1000  ∞

Limits in AI

LimitValueMeaning in AIlim(x→∞) sigmoid(x)1Output saturates at 1 (confident positive)lim(x→-∞) sigmoid(x)0Output saturates at 0 (confident negative)lim(x→0) ln(x)-∞Loss explodes for wrong predictionslim(n→∞) (1+1/n)ⁿeFoundation of all exponentialslim(lr→0) gradient stepexact derivativeSmaller learning rate = more precise

Key Takeaway

lim tells you where a function is HEADING,
even if it never actually arrives there.

In AI:
  - Sigmoid APPROACHES 0 and 1 but never reaches them
  - Loss APPROACHES infinity as prediction → 0
  - e IS the limit of (1+1/n)ⁿ as n → ∞
  - Derivative IS the limit of (f(x+h)-f(x))/h as h → 0