What is a curve doing right now?

Δ over Δ — average rate

You drove from $x = 1 km$ at $t = 0 s$ to $x = 9 km$ at $t = 4 s$. Your average speed is $(9 − 1) / (4 − 0) = 2 km/s$. That number describes the trip as a whole — but it does not tell you what you were doing at $t = 2 s$. The line through $(0, 1)$ and $(4, 9)$ on the position graph is a secant , and its slope is the average. You can compute it without any limit.

# Average rate of change — the secant slope. Two points, one division.
def average_rate(f, a, b):
    return (f(b) - f(a)) / (b - a)

f = lambda t: t * t            # x(t) = t² — toy "position"
average_rate(f, 1, 3)          # → 4.0   (position went 1 → 9 in 2 seconds)

Shrink the interval — instantaneous rate

“What was I doing at $t = 2 s$?” requires the secant’s two points to merge into one. Pick a fixed anchor $a = 2$ and a small interval $h$; compute $(f(a+h) − f(a)) / h$ as $h$ shrinks. For $f(t) = t²$:

(f(2+h) − f(2)) / h
 = ((2+h)² − 4) / h
 = (4 + 4h + h² − 4) / h
 = 4 + h            ← independent of how big h is, except for the +h tail

As $h → 0$, the expression converges to $4$. Not “approaches” in some hand-wavy sense — the value is $4 + h$, and $h$ can be made as small as you want. The number that survives in the limit, 4, is the derivative of $t²$ at $t = 2$: $f'(2) = 4$. Geometrically, the slope of the tangent line through $(2, 4)$.

# Instantaneous rate — shrink the interval and watch the secant slope
# converge to the tangent slope. No epsilon-delta; just shrink and look.
def secant_slope(f, a, h):
    return (f(a + h) - f(a)) / h

[secant_slope(f, 3, h) for h in (1, 0.1, 0.01, 0.0001)]
# → [7.0, 6.1, 6.01, 6.0001]
# The pattern: 6 + h. The limit as h → 0 is 6.
# That's f'(3) for f(t) = t².  In general, f'(t) = 2t.

One machine, three names — slope, velocity, rate

The recipe — pick two points, compute the secant slope, shrink the interval, take the limit — gives the same kind of number whatever you plug in. If $f$ is a position-vs-time function, the derivative is velocity . If $f$ is a graph of revenue vs price, the derivative is “marginal revenue.” If $f$ is a curve drawn on paper, the derivative is the tangent’s slope at each point. Same operation; different physical or geometric interpretation depending on what was on the axes.

The standard pattern: derivative of $x^n$ is $n·x^(n−1)$. The widget shows it for $n = 2$: drag the anchor and read off $2a$. The same machinery, applied to $\sin x$, gives $\cos x$; applied to $e^x$, gives $e^x$ back; applied to a constant, gives 0. The names of these — “differentiation rules” — are bookkeeping. The single underlying operation is the limit of secant slopes.

Differentiate twice — acceleration

The derivative of a function is itself a function. You can differentiate it again. For $f(t) = t²$: $f'(t) = 2t$ (a line); $f''(t) = 2$ (a constant). Two derivatives of position give acceleration — the rate of change of velocity, which for free fall is the constant $−g$.

That tower — position, velocity, acceleration — is the entire content of “Newton’s second law” once you have the derivative as a tool: force is mass times the second derivative of position. Most introductory physics is the algebraic and geometric consequences of this single fact.

Where this shows up — same operation, two pillars

A derivative is not just a slope. It is local change: how one quantity responds when another is nudged. The same operation shows up under different names in different pillars; the algebra stays the same.

physics : position changes into velocity; velocity into acceleration;
        forces decide those changes.
ml      : loss changes when a parameter moves; the gradient tells
        which way the loss falls.
finance : a price changes when a rate moves; the derivative measures
        the sensitivity — duration for bonds, the Greeks for options.

Projectile motion — the equations $x(t) = v₀ \cos θ · t$ and $y(t) = v₀ \sin θ · t − \tfrac{1}{2} g t²$ have derivatives $vₓ = v₀ \cos θ$ (constant) and $v_y(t) = v₀ \sin θ − g t$ (linear). Differentiating the position gives the velocity directly; one more derivative gives the constant $−g$ acceleration.

The pendulum clock — the equation of motion $\ddot{θ} = −(g/L) \sin θ$ is two derivatives of θ on the left, equals a function of θ on the right. Replace $\sin θ$ with $θ$ (the linearization trick) and you get $\ddot{θ} = −(g/L) θ$ — simple harmonic motion , which the derivative tool can recognize and solve.

Terminal velocity — the differential equation $dv/dt = g − k v$ is one derivative on the left, the net force per unit mass on the right. The fixed point — the terminal speed — is found by setting the derivative to zero: $g − k v_t = 0$. Force balance is exactly “the derivative vanishes here.”

Damped oscillator — the equation $\ddot{x} + 2γ\dot{x} + ω₀² x = F(t)$ runs two derivatives on the left: a velocity term and an acceleration term. Damping is exactly the term that multiplies the first derivative; resonance is the place where forcing matches the second-derivative geometry. The whole page is the derivative ladder, twice.

Gradient descent — the loss $L(w)$ is a multivariable function; its gradient $∇L$ is the vector of partial derivatives. The descent step $w ← w − η · ∇L$ uses one derivative per parameter, all at once. The rate at which the loss falls is $\|∇L\|²$ — a derivative-derived quantity that determines whether learning is making progress.

Present value — the same operation, in finance vocabulary. The price of a bond is $PV(r) = \int_0^T c(t) · e^{-rt} dt$, and the modified duration $D = -(\partial PV / \partial r) / PV$ is a derivative, normalized — it predicts how much the bond price moves when rates shift. Option Greeks are the same family: $\Delta = \partial \text{Price} / \partial S$, $\Gamma$ the second derivative, and so on. Financial derivative and mathematical derivative are the same word for a reason.

Same idea, different nouns: velocity in physics, gradient in ML, sensitivity in finance. In all three the derivative is the local rule for change.

# Position → velocity → acceleration. Same machine, applied twice.
# Projectile: y(t) = v₀ sin θ · t − ½ g t²    (from /physics/projectile-motion)
#   dy/dt = v₀ sin θ − g t                    (vertical velocity)
#   d²y/dt² = −g                              (vertical acceleration — constant)
#
# Pendulum (small-angle): θ(t) = θ₀ cos(ω t),  ω = √(g/L)
#   dθ/dt  = −θ₀ ω sin(ω t)
#   d²θ/dt² = −θ₀ ω² cos(ω t) = −ω² · θ(t)   ← simple harmonic motion
#
# Each application's equation of motion is one or two derivatives applied
# to the position function. The derivative is the shared tool.