Function Transformations

Function transformations modify a function's graph through translations, reflections, stretches, and compressions by altering its input or output. Key result: changes outside the function (

a f (x) + k

) affect outputs vertically, while changes inside (

f (b x + c)

) affect inputs horizontally and behave "backwards." They apply anywhere graphs of functions appear — physics, signal processing, statistics, computer graphics. Systematized in modern form through 19th–20th-century function theory and standardized in 20th-century algebra curricula.

A function transformation takes a known "parent" function $f (x)$ and produces a new function whose graph is a geometric modification of the original. The four basic operations are translation (shifting), reflection (flipping), stretching (scaling away from an axis), and compression (scaling toward an axis). These matter because complex graphs can be understood as simple parents reshaped by predictable rules, turning graphing into a systematic process rather than point-by-point plotting. The surprising fact that trips up nearly every student: horizontal transformations operate inversely to what the algebra appears to say — $f (x + 3)$ shifts the graph left, not right.

The Core Insight

Every transformation is either an output operation (applied to the whole function $f (x)$ , affecting $y$ -values, behaving intuitively) or an input operation (applied to $x$ before $f$ acts, affecting $x$ -values, behaving inversely). Master that single inside/outside distinction and every rule follows. The general transformed function

g (x) = a f (b (x - h)) + k

⟳Compose the parent

f

with an inner linear map on the input and an outer linear map on the output; each constant

a, b, h, k

controls exactly one geometric effect.

contains all standard transformations as special cases.

Formal Statement

The Four Transformation Families

Let $f$ be a function ("parent") and let $a, b, h, k$ be real constants. The transformed function $g (x) = a f (b (x - h)) + k$ decomposes as follows, where each symbol controls one geometric action:

$k$ — vertical translation: shifts the graph up ( $k > 0$ ) or down ( $k < 0$ ) by $∣ k ∣$ units.
$h$ — horizontal translation: shifts the graph right ( $h > 0$ ) or left ( $h < 0$ ) by $∣ h ∣$ units.
$a$ — vertical scaling: stretches ( $∣ a ∣ > 1$ ) or compresses ( $0 < ∣ a ∣ < 1$ ) vertically by factor $∣ a ∣$ ; reflects across the $x$ -axis if $a < 0$ .
$b$ — horizontal scaling: compresses ( $∣ b ∣ > 1$ ) or stretches ( $0 < ∣ b ∣ < 1$ ) horizontally by factor $1/∣ b ∣$ ; reflects across the $y$ -axis if $b < 0$ .

The individual rules, isolated:

g (x) = f (x) + k ⟹ (x, y) \mapsto (x, y + k)

⟳Adding

k

outside changes every output by the same amount, leaving inputs untouched.

g (x) = f (x - h) ⟹ (x, y) \mapsto (x + h, y)

⟳A point reaches the same output value

y = f (x_{0})

only when the new input satisfies

x - h = x_{0}

, i.e.

x = x_{0} + h

, so the graph moves

+ h

g (x) = a f (x) ⟹ (x, y) \mapsto (x, a y)

⟳Multiplying outputs by

a

scales vertical distances from the

x

-axis by

a

; negative

a

flips the sign, reflecting across the

x

-axis.

g (x) = f (b x) ⟹ (x, y) \mapsto (\frac{x}{b}, y)

⟳To feed

f

the same internal value

x_{0}

, set

b x = x_{0}

, so

x = x_{0} / b

; inputs scale by

1/ b

, the inverse of the multiplier.

Order of Composition

For the general form $g (x) = a f (b (x - h)) + k$ , applying transformations to a point $(x_{0}, f (x_{0}))$ in the order that respects function composition gives the image point

(\frac{x _{0}}{b} + h, a f (x_{0}) + k) .

⟳Read the formula from the inside out for inputs (horizontal scale

1/ b

, then shift

+ h

) and from outside in for outputs (vertical scale

a

, then shift

+ k

); horizontal and vertical chains are independent.

Order Matters Within an Axis

Along a single axis, scaling and shifting do not commute. Because $h$ sits inside the scaling in form [1], the correct horizontal procedure is: scale by $1/ b$ first, then translate by $h$ . If a problem is written as $f (b x - c)$ , you must factor it as $f (b (x - c / b))$ before reading off the shift — the shift is $c / b$ , not $c$ .

Why It Holds / Derivation

The Inside/Outside Principle

We justify the inverse behavior of horizontal transformations, the single most error-prone rule.

Step 1. Let

(x_{0}, y_{0})

be a point on the parent graph, so

y_{0} = f (x_{0})

. Step 2. Consider

g (x) = f (x - h)

. We ask: for which input

x

does

g

output the same height

y_{0}

? Step 3. We need

f (x - h) = f (x_{0})

, which holds when the internal argument matches:

x - h = x_{0}

. Step 4. Solving gives

x = x_{0} + h

. So the point that had

x

-coordinate

x_{0}

on the parent now sits at

x_{0} + h

g

— the graph shifted right by

h

(for

h > 0

). This is why

f (x - h)

, which looks like subtraction, moves the graph in the positive direction. Step 5. For scaling

g (x) = f (b x)

: matching internal arguments

b x = x_{0}

gives

x = x_{0} / b

. Each

x

-coordinate is divided by

b

, so a multiplier

b > 1

pulls points toward the

y

-axis (compression).

■

Think of the parent graph as already "decided" what output it gives for each internal value. Outside operations ( $a, k$ ) edit the answer after the function speaks — so they behave exactly as written. Inside operations ( $b, h$ ) edit the question before the function hears it — so to keep the same answer, $x$ must compensate in the opposite direction. The function is "deaf to delay": to make it respond at a new place, you feed it the old internal value, and the algebra inverts.

Joseph Fourier | 1768–1830

While transformations as a pedagogical unit are recent, Fourier's analysis of heat conduction (1822) made scaling and shifting of functions a central computational tool. Decomposing signals as sums of stretched and shifted sinusoids — $A sin (b (x - h)) + k$ — turned function transformations from graphical curiosities into the workhorse of physics and engineering, embedding amplitude $A$ , frequency $b$ , phase shift $h$ , and vertical offset $k$ as the four parameters every student now meets.

Building Intuition

The four transformations of the parent parabola $f(x)=x^2$

Worked Examples

Transforming a Square-Root Function

Given: Parent

f (x) = x

. Find

g (x) = - 2 x - 3 + 1

and describe its transformations and the image of the parent's start point

(0, 0)

. Identify constants:

a = - 2

b = 1

h = 3

k = 1

. Step 1 — Vertical scale & reflect:

a = - 2

stretches vertically by factor

2

and reflects across the

x

-axis. Step 2 — Horizontal shift:

h = 3

moves the graph right

3

units. Step 3 — Vertical shift:

k = 1

moves the graph up

1

unit. Step 4 — Track the start point

(0, 0)

using formula [6] with

b = 1

x : 0/1 + 3 = 3, y : - 2 (0) + 1 = 1.

Answer:

g (x) = - 2 x - 3 + 1; start point (0, 0) \to (3, 1)

The new curve begins at

(3, 1)

and falls to the right (opening downward due to reflection).

A Sinusoid with All Four Parameters

Given: Parent

f (x) = sin x

. Analyze

g (x) = 3 sin (2 x - \frac{π}{2}) + 4

. Step 1 — Factor inside for the true shift: rewrite the argument as

2 (x - \frac{π}{4})

. So

b = 2

h = \frac{π}{4}

. Step 2 — Read constants:

a = 3

(amplitude),

b = 2

h = \frac{π}{4}

(phase shift right),

k = 4

(midline). Step 3 — Compute features:

Amplitude $= ∣ a ∣ = 3$ .
Period $= \frac{2 π}{∣ b ∣} = \frac{2 π}{2} = π$ .
Phase shift $= \frac{π}{4}$ to the right.
Midline $y = 4$ ; range $[4 - 3, 4 + 3] = [1, 7]$ .

Step 4 — Image of the parent peak

(\frac{π}{2}, 1)

via [6]:

x : \frac{π /2}{2} + \frac{π}{4} = \frac{π}{4} + \frac{π}{4} = \frac{π}{2}, y : 3 (1) + 4 = 7.

Answer:

Amplitude 3, period π, phase shift \frac{π}{4} right, midline y = 4; peak at (\frac{π}{2}, 7)

Common Misconceptions

The Five Classic Errors

Direction of horizontal shift: $f (x + 3)$ shifts left, $f (x - 3)$ shifts right — opposite to the sign you "see."
Forgetting to factor: In $f (2 x - 6)$ the horizontal shift is $3$ (factor to $f (2 (x - 3))$ ), not $6$ .
Confusing stretch and compression horizontally: $f (2 x)$ compresses by factor $\frac{1}{2}$ ; $f (\frac{x}{2})$ stretches by factor $2$ — again inverted.
Mixing up which factor is vertical: $a$ (outside) is vertical; $b$ (inside) is horizontal.
Wrong order within an axis: scale before shifting horizontally when the form is $a f (b (x - h)) + k$ .

Stretch ≠ Shift Equivalence

A student claims $g (x) = (2 x)^{2}$ is "the parabola $x^{2}$ shifted." It is not a shift at all — it equals $4 x^{2}$ , a vertical stretch by 4 (equivalently a horizontal compression by $\frac{1}{2}$ ). Test points confirm: parent at $x = 1$ gives $1$ ; $g (1) = 4$ , but a horizontal shift would preserve the output value $1$ somewhere. No translation can change the height of a vertex while keeping it on the axis — only scaling does that. This shows scaling and translation are genuinely distinct operations, not interchangeable.

Connections

Function transformations sit at a crossroads of several topics. They are the graphical face of Function Composition — every transformation is $f$ composed with linear maps inside and out. The horizontal/vertical inversion is illuminated by Inverse Functions, where reflecting across $y = x$ swaps the roles of input and output transformations entirely. In trigonometry, transformations parameterize the general Sinusoidal Functions via amplitude, period, and phase. They also prefigure Linear Transformations in linear algebra, where scaling and shifting generalize to matrices and vectors, and they underlie the study of Even and Odd Functions, whose symmetry is precisely invariance under reflection transformations.

Real-World Applications

Audio engineering: A volume knob applies vertical scaling $a$ ; a pitch/time shift applies horizontal scaling $b$ to a sound's waveform $f (t)$ .
Seismology: Earthquake waveforms are amplitude-scaled and time-shifted to align recordings from different stations and locate epicenters.
Computer graphics & animation: Sprite movement is horizontal/vertical translation; zooming is scaling; mirroring a character uses reflection across an axis.
Economics (supply/demand): A tax shifts a supply curve vertically by $k$ ; a change in market size scales demand horizontally.
Physics — simple harmonic motion: Position $x (t) = A cos (ω t - φ) + x_{0}$ encodes amplitude $A$ , angular frequency $ω$ (horizontal scale), phase $φ$ , and equilibrium offset $x_{0}$ .
Statistics: Standardizing data, $z = \frac{x - μ}{σ}$ , is a horizontal shift by $μ$ and scale by $1/ σ$ applied to the normal curve.
Image processing: Brightness adjustment adds $k$ (vertical shift); contrast multiplies by $a$ (vertical scale) on the pixel-intensity function.

Historical Development

c. 1637 — Descartes' La Géométrie introduces coordinate geometry, making it possible to graph algebraic relations and thus to compare transformed curves.
1748 — Euler's Introductio in analysin infinitorum establishes the modern concept of a function as the object that can be transformed.
1822 — Fourier represents heat distributions as sums of scaled, shifted sinusoids, embedding the four parameters into physics.
1872 — Felix Klein's Erlangen Program recasts geometry as the study of figures invariant under transformation groups, giving transformations a unifying theoretical home.
1893 — Sophus Lie's work on continuous transformation groups formalizes scaling and translation as group actions.
1900s–1920s — Standardized secondary-school algebra texts in the US and Europe begin presenting translations and reflections of conics explicitly.
1950s–1960s — The "New Math" movement and function-centered curricula (influenced by Bourbaki) make $y = a f (b (x - h)) + k$ a standard classroom template.
1989 — NCTM Curriculum and Evaluation Standards in the US enshrines transformations of parent functions as core content, cementing today's textbook treatment.

Subtleties & Limits

Where the Simple Rules Strain

The clean four-parameter model assumes transformations act on a single variable independently. Subtleties arise when: (1) non-affine inner functions appear — e.g. $f (x^{2})$ is not a simple shift/scale and warps the domain nonuniformly; (2) transformations interact with domain restrictions — shifting $ln (x)$ left can push part of the graph into undefined territory; (3) multivariable generalizations require matrix transformations where scaling and rotation no longer commute and "horizontal vs. vertical" loses meaning. The deeper question — which transformations preserve which structural properties of a function (continuity, periodicity, symmetry, integrability) — opens into functional analysis and the theory of operators, where "transformation" becomes a linear operator on infinite-dimensional function spaces, far beyond the introductory picture.