Calculus I: Functions, Limits, Derivatives and Integrals

The Mathematics I progression follows the undergraduate Computer Engineering sequence used in 2013-2014: functions, limits, differentiation, and integration. Definitions, assumptions, and notation are presented with emphasis on the conditions under which each result is valid.

Being able to execute a calculation is not the same as knowing the assumptions behind it. Limits, derivatives and integrals therefore form one analytical chain, with symbolic and numerical interpretations kept separate where their error mechanisms differ.

Unit 1: Analytic Prerequisites and Coordinate Geometry

This section collects the algebraic and geometric background that calculus uses continuously but often leaves implicit. The goal is not to reproduce a separate course in set theory or analytic geometry; it is to keep the notation and geometric meaning required by functions, limits, derivatives, and integrals explicit.

1.1. The Real Line, Intervals, and Absolute Value

Real numbers are ordered on the number line. Open and closed interval notation records whether an endpoint belongs to the set:

(a,b)   = {x ∈ R : a < x < b}
[a,b]   = {x ∈ R : a ≤ x ≤ b}
[a,b)   = {x ∈ R : a ≤ x < b}
(a,b]   = {x ∈ R : a < x ≤ b}

Infinity is not a real endpoint, so an infinite side is always written as open:

(-∞,a),   [a,∞),   (-∞,∞)

Absolute value is distance on the real line:

|x-a| = distance between x and a

This interpretation appears directly in the epsilon-delta definition of a limit. For example,

|x-a| < r

means that x lies in the open interval of radius r centered at a:

a-r < x < a+r

Reading an absolute-value inequality as a distance constraint often makes the geometry clearer than treating it only as a symbolic manipulation.

1.2. Distance and Midpoint in the Cartesian Plane

For P(x1,y1) and Q(x2,y2), the Euclidean distance follows from the Pythagorean relation:

d(P,Q) = √[(x2-x1)^2 + (y2-y1)^2]

The midpoint is the coordinatewise arithmetic mean:

M = ((x1+x2)/2, (y1+y2)/2)

Both formulas reappear later in arc length, norms, minimum-distance problems, and geometric optimization.

1.3. Slope and Forms of a Line

When x1 ≠ x2, the slope of the line through two points is

m = (y2-y1)/(x2-x1).

Slope measures the change in y produced by a unit change in x. The derivative can therefore be viewed as the local limiting form of this average slope.

Common line equations are

Point-slope:       y-y0 = m(x-x0)
Slope-intercept:   y = mx+b
General form:      Ax+By+C = 0

A vertical line has the form x=c and has no finite slope. Saying that its slope is undefined is more precise than assigning infinity as a real numerical value.

Parallel nonvertical lines have equal slopes. For finite nonzero slopes,

m1 · m2 = -1

characterizes perpendicularity. The normal-vector formulation is more general and extends naturally to three dimensions; that viewpoint is developed in Linear Algebra.

1.4. Analytic Representation of a Circle

A circle with center (a,b) and radius r > 0 satisfies

(x-a)^2 + (y-b)^2 = r^2.

For an equation of the form

x^2 + y^2 + Dx + Ey + F = 0,

completing the square recovers the center-radius form. The resulting r^2 determines the real geometry:

• r^2 > 0: a real circle,
• r^2 = 0: a single point,
• r^2 < 0: no real point satisfies the equation.

This classification matters when establishing domains, intersections, and integration regions.

1.5. Quadratic Equations and the Geometric Meaning of the Discriminant

For

ax^2 + bx + c = 0,   a ≠ 0,

the discriminant

Δ = b^2 - 4ac

separates the real-root cases. Δ > 0 gives two distinct real roots, Δ = 0 a repeated real root, and Δ < 0 no real root.

In calculus the discriminant often appears indirectly when finding intersections, sign intervals, or integration limits. In area problems, the boundaries should be obtained from the actual intersections rather than assumed from a sketch.

1.6. Boundary Between Prerequisites and Later Courses

The material here is used as infrastructure for calculus. Formal set operations belong primarily to Discrete Mathematics, vector spaces and affine geometry to Linear Algebra, and approximate root finding to Numerical Analysis. Keeping those ownership boundaries prevents the same topic from being expanded independently in several courses.

Unit 2: Functions

2.1. Function Concept

A function f:A→B assigns exactly one value f(x) in the codomain B to every x in the domain A. The domain, codomain, and rule together determine the function.

2.2. Determining the Domain

For an expression used to define a real-valued function, the domain excludes points where the expression is undefined. Typical restrictions arise from division by zero, even roots of negative values, and logarithms of nonpositive quantities.

2.3. Range, into and Onto Functions

The range is the set of values actually attained by the function. A function is onto/surjective when its range equals the codomain. An into function leaves at least one codomain value unattained.

2.4. One-to-one Function, Identity Function, Equal Functions

A function is one-to-one/injective if

f(x1)=f(x2) ⇒ x1=x2.

The identity function satisfies I(x)=x. Two functions are equal when their domains, codomains, and values agree.

2.5. Arithmetic Operations on Functions

For functions with compatible domains,

(f±g)(x)=f(x)±g(x)
(fg)(x)=f(x)g(x)
(f/g)(x)=f(x)/g(x),  g(x)≠0.

The resulting domain is the intersection of the relevant domains together with any additional restrictions.

2.6. Composite Function

(g∘f)(x)=g(f(x)).

The value f(x) must lie in the domain of g. Composition is associative but not generally commutative.

2.7. Inverse Function

A function has an inverse on its stated domain and codomain when it is bijective. Then

f^-1(f(x))=x
f(f^-1(y))=y.

An injective restriction of a function can sometimes be used to define an inverse on a smaller domain.

2.8. Graph of a Function

The graph of y=f(x) is the set of points (x,f(x)). A vertical line intersects the graph of a real function of one variable at most once.

2.9. Even and Odd Functions

On a domain symmetric about zero,

f(-x)=f(x)

defines an even function, while

f(-x)=-f(x)

defines an odd function. Their graphs have y-axis and origin symmetry respectively.

2.10. Bounded and Unbounded Functions

A function is bounded above if f(x)≤M for some M and all domain points, bounded below analogously, and bounded when both hold.

2.11. Maximum, Minimum, Supremum, and Infimum

A maximum or minimum must be attained by the function. The supremum is the least upper bound and the infimum the greatest lower bound; either can exist without being attained.

2.12. Monotone Functions

A function can be increasing/decreasing or strictly increasing/decreasing on an interval. Monotonicity provides information about injectivity and later about derivatives and extrema.

2.13. Periodic Functions

A function is periodic if there exists T>0 such that

f(x+T)=f(x)

throughout the relevant domain. The least positive such T, when it exists, is the fundamental period.

2.14. Absolute-value and Sign Functions

Absolute value is

|x| = x   if x≥0
     = -x if x<0.

The sign function records whether a real number is positive, zero, or negative.

2.15. Exponential and Logarithmic Functions

For a>0, a≠1, the exponential a^x is positive and one-to-one. The logarithm log_a x is its inverse on positive numbers. Their derivatives and integrals form a central part of calculus.

2.16. Exponential Growth, Decay, and Compounding

The defining modeling property of an exponential function is that each step changes the current quantity by a fixed proportion, rather than by a fixed absolute amount. If a quantity grows by rate r per period from an initial value A₀, then after n periods

Aₙ = A₀(1 + r)ⁿ.

The same structure models exponential decay when r < 0. The assumption is that the proportional rate remains constant across periods; if the rate depends on time or state, the model must change accordingly.

When time is treated as continuous, the model can be written as

A(t) = A₀ e^(kt),

where k > 0 gives growth and k < 0 gives decay. For an equivalent one-period growth factor, the discrete and continuous rates satisfy

k = ln(1 + r).

The exponential and logarithm therefore act not only as inverse functions but also as transformations between equivalent growth-rate representations.

This structure appears in population models, decay processes, compound growth, error attenuation, reliability models, and selected feedback systems. A mathematically valid exponential form does not by itself establish that observed data are exponential; parameter estimation and model validation remain separate tasks.

2.17. Cost, Revenue, Profit, and break-even Points

Functions can model quantitative relationships within a system. Let x denote production quantity, C(x) total cost, and p(x) unit selling price. Total revenue is

R(x) = x · p(x),

and profit is

Π(x) = R(x) − C(x).

If the system has a fixed initial cost and a constant variable cost per unit, then

C(x) = C₀ + c·x

is a linear cost model. Revenue need not be linear when price depends on quantity. If p(x) decreases with quantity, for example, R(x)=x·p(x) is generally nonlinear.

Values satisfying

Π(x) = 0

are break-even points. They can be interpreted either as algebraic roots of the profit function or as intersections of the revenue and cost functions.

The mathematical domain must also be restricted by the physical model. Negative production, output beyond capacity, or price ranges outside the model's validity may be algebraically admissible yet infeasible. This distinction separates a formal solution set from the set of operationally meaningful solutions.

2.18. Hyperbolic Functions and Inverses

Hyperbolic sine and cosine are defined from exponentials:

sinh x = (e^x-e^-x)/2
cosh x = (e^x+e^-x)/2.

They satisfy identities analogous to, but distinct from, trigonometric identities.

2.19. Power Functions and Polynomials

Power functions have the form x^α on an appropriate real domain. Polynomials are finite sums of nonnegative integer powers and are continuous and differentiable everywhere on R.

2.20. Basic Transformations for Curve Sketching

Translations, scalings, reflections, and compositions transform known graphs. Expressions such as f(x-a)+b, cf(x), f(cx), and -f(x) have direct geometric interpretations.

2.21. Reading Function, Change, and Accumulation in One Mathematical Framework

The first course in calculus is organized around local and accumulated change:

Functions
   ↓
Limits
   ↓
Continuity
   ↓
Derivatives
   ↓
Local approximation and qualitative behavior
   ↓
Antiderivatives
   ↓
Definite integrals and accumulation
   ↓
Improper limits and numerical approximation

Limits provide the common language beneath continuity, derivatives, and integrals. Derivatives describe local rate; integrals describe accumulation. The fundamental theorem connects the two operations under appropriate hypotheses.

2.22. Boundaries Among Limits, Continuity, Derivatives, and Integrals

Domain ≠ range. The domain contains admissible inputs; the range contains values actually produced.

Injective ≠ surjective. Injectivity prevents two distinct inputs from sharing one output; surjectivity requires every codomain element to be reached.

Maximum ≠ supremum. A maximum is attained; a supremum need only be the least upper bound.

Function value ≠ limit. f(a) can differ from lim_(x→a)f(x) or even be undefined while the limit exists.

Infinite limit ≠ finite real limit. It describes unbounded behavior rather than convergence to a real number.

Continuity ≠ differentiability. Differentiability implies continuity; continuity does not imply differentiability.

Critical point ≠ extremum. f'=0 or failure of differentiability identifies candidates; additional analysis is needed.

f''(a)=0 ≠ inflection point. Concavity must actually change.

Differential ≠ exact finite change. dy=f'(x)dx is the first-order linear part of the change.

Indefinite integral ≠ definite integral. The first is a family of antiderivatives; the second is a number defined by an accumulation limit.

Signed integral ≠ geometric area. Negative portions subtract in a signed integral.

Improper integral ≠ ordinary integral with infinity as a number. It is defined through a limit.

The durable outcome of Calculus I is not a catalogue of derivative and integral patterns. It is the ability to connect local change with accumulated effect through the language of limits. That same distinction reappears later in numerical approximation, convergence analysis, simulation, and continuous-time engineering models.

2.23. Read the Function Before Computing

Derivative and integral questions should begin with the domain and behaviour of the function. If the function is undefined at a point, a derivative at that point cannot be assumed. Existence of a limit does not imply that the function value exists. Continuity requires the limit to exist, the function value to be defined, and the two to agree.

A derivative is a local rate of change. The chain rule preserves dependency in a composite function:

d/dx f(g(x)) = f'(g(x)) · g'(x)

In optimisation, f'(x)=0 identifies candidate stationary points; it does not by itself prove a maximum or minimum. Endpoints, nondifferentiable points, and either sign changes or second-derivative information may also matter.

A definite integral represents signed accumulation under suitable assumptions. Regions below the axis contribute negatively; geometric area may therefore require absolute values or splitting the interval. The fundamental theorem connects differentiation and integration but does not imply that every antiderivative has an elementary closed form.

Improper integrals handle infinite bounds or singularities through limits. Writing the integral symbol is not evidence of convergence; if the defining limit is not finite, the integral diverges.

Quick dimensional and qualitative checks are useful. A derivative changes units roughly as y/x, an integral as y·x; the derivative sign should match local increase or decrease, and a definite integral should be compatible with the rough geometric scale of the function.

Unit 3: Limit of a Function

To evaluate a limit, attempt direct substitution and identify the resulting expression or indeterminate form. Choose algebraic simplification, rationalization, or standard-limit methods according to the expression's structure. Inspect approaches from the left and right separately; unequal one-sided limits exclude a two-sided limit. During algebraic transformations, remember that points where a denominator vanishes remain excluded from the transformed expression's domain.

3.1. Limit Concept

The statement

lim_(x→a) f(x)=L

describes values of f(x) near a, not necessarily the value f(a) itself.

3.2. Epsilon-delta Definition

For every ε>0, there must exist δ>0 such that

0<|x-a|<δ ⇒ |f(x)-L|<ε.

This definition makes the informal phrase "arbitrarily close" precise.

3.3. Right- and left-hand Limits

lim_(x→a+) f(x)
lim_(x→a-) f(x)

consider approaches from one side. A finite two-sided limit exists exactly when both one-sided limits exist and are equal.

3.4. Infinite Limit at a Finite Point and Vertical Asymptote

If |f(x)| grows without bound as x approaches a from an appropriate side, the function has an infinite limit there. The line x=a is then a vertical asymptote in the corresponding sense.

3.5. Limit at Infinity and Horizontal Asymptote

If

lim_(x→∞) f(x)=L

or the analogous limit at -∞ exists, y=L is a horizontal asymptote on that side.

3.6. Properties of Limits

Under the usual existence conditions, limits respect sums, products, scalar multiples, quotients with nonzero limiting denominator, and continuous compositions.

3.7. Limit of a Monotone Function

Monotone functions have one-sided limits at interior points in the extended-real sense. Bounded monotone behavior also underlies convergence results for sequences.

3.8. Trigonometric Limits

A fundamental limit is

lim_(x→0) sin x / x = 1,

with angles measured in radians. It supports the derivative formulas for sine and cosine and many related limits.

3.9. Indeterminate Forms

Forms such as

0/0,  ∞/∞,  0·∞,  ∞-∞,  0^0,  1^∞,  ∞^0

are not values. They signal that more analysis is required.

3.10. Selected Limit Rules

Algebraic simplification, rationalization, factorization, standard limits, substitutions, squeeze arguments, series, or L'Hôpital's rule under its hypotheses can resolve different limit forms.

Unit 4: Continuous Functions

4.1. Continuity Concept

A function is continuous at a if

lim_(x→a) f(x)=f(a).

This requires f(a) to exist, the limit to exist, and the two to be equal.

4.2. Increment Definition of Continuity

With Δx→0, continuity can be expressed as

Δf = f(a+Δx)-f(a) → 0.

4.3. Right- and left-continuity

At endpoints or piecewise definitions, one-sided continuity uses the corresponding one-sided limit.

4.4. Discontinuities of the First and Second Kind

A discontinuity is often called first kind when both finite one-sided limits exist, including removable and jump discontinuities. More severe behavior, such as an infinite or nonexisting one-sided limit, is classified as second kind in the classical terminology used by the course.

4.5. Properties of Continuous Functions

Sums, products, and suitable quotients of continuous functions remain continuous. Compositions of continuous functions are continuous where defined.

4.6. Continuous Functions on Closed Intervals

A function continuous on [a,b] is bounded and attains a maximum and minimum. The intermediate value theorem guarantees that it takes every value between f(a) and f(b).

Unit 5: Derivatives and Differentials

5.1. Derivative Concept

The derivative at x is

f'(x)=lim_(h→0) [f(x+h)-f(x)]/h

when this limit exists.

5.2. Differentiability and Continuity

Differentiability at a point implies continuity there. The converse is false; a continuous function can have a corner, cusp, or other nondifferentiable behavior.

5.3. Geometric Meaning of the Derivative

For a graph y=f(x), f'(a) is the slope of the tangent line when the ordinary finite derivative exists:

y-f(a)=f'(a)(x-a).

5.4. Physical Meaning of the Derivative

If s(t) is position, s'(t) is instantaneous velocity and s''(t) acceleration. More generally, derivatives model instantaneous rates of change.

5.5. Marginal Analysis and Finite Differences

A derivative measures a local rate of change, whereas an observed change in a discrete system is often a finite difference. For a cost function,

C(x + Δx) − C(x)

is the change in total cost when production increases by Δx. The average rate of change is

[C(x + Δx) − C(x)] / Δx,

which approaches C'(x) as Δx → 0.

For a one-unit increment,

C(x + 1) − C(x) ≈ C'(x)

can be useful, but it is not an identity. Under the hypotheses of the mean value theorem, there is some ξ between x and x+1 such that

C(x + 1) − C(x) = C'(ξ).

The closeness of C'(x) to the actual one-unit change therefore depends on how much the derivative varies across that interval.

The same reasoning is not specific to economics. The latency effect of adding one request to a queue, the output effect of changing sensor resolution, or the energy effect of increasing system capacity can all be studied through the same local-sensitivity idea.

5.6. Elasticity and Dimensionless Sensitivity

The raw derivative

f'(x)

measures local output response to a one-unit input change, but its numerical value depends on units and scale. Relative change can be described with elasticity, or normalized sensitivity:

E_f(x) = [x / f(x)] · f'(x).

When x and f(x) are nonzero, this can be read as

E_f(x) = (df/f) / (dx/x).

It approximates the percentage output change associated with a one-percent input change. Because it is dimensionless, it is invariant to unit changes that merely rescale the same quantity.

For a power law

f(x) = a·xⁿ,

the elasticity is

E_f(x) = n.

Thus the exponent is also a constant relative sensitivity.

For decreasing functions such as demand curves, elasticity may be negative. Some disciplines report its magnitude |E|; whether the sign is retained is a convention that should be stated explicitly.

Elasticity shares the same normalization idea as relative error and conditioning in Numerical Analysis: when absolute change alone is misleading, change is evaluated relative to the scale of the quantity itself.

5.7. One-sided and Infinite Derivatives

One-sided derivatives use h→0+ or h→0-. If slopes grow without bound, a vertical-tangent type behavior may be described through an infinite derivative, although this is distinct from having an ordinary finite derivative.

5.8. Differential of a Function

For a differentiable function,

dy = f'(x) dx.

For a small increment, dy is the linear part of the change in f and provides local approximation.

5.9. Basic Differentiation Rules

(c)'=0
(x^n)'=n x^(n-1)
(f+g)'=f'+g'
(fg)'=f'g+fg'
(f/g)'=(f'g-fg')/g^2.

5.10. Derivative of a Composite Function

The chain rule is

(g∘f)'(x)=g'(f(x)) f'(x).

The chain rule can also be understood as the composition of two local changes. With u = g(x) and y = f(u), a small perturbation Δx produces Δu ≈ g'(x₀)Δx, followed by Δy ≈ f'(u₀)Δu. Combining these approximations gives the geometric meaning of the derivative formula above. If the inner map locally scales change by 2 and the outer map by 1.5, the composite derivative is 2 × 1.5 = 3; a negative derivative reverses the local direction.

5.11. Derivative of an Inverse Function

If f is locally invertible and f'(x)≠0, then

(f^-1)'(y)=1/f'(x),  y=f(x).

5.12. Derivatives of Elementary Functions

Standard derivative formulas include exponential, logarithmic, trigonometric, inverse-trigonometric, and hyperbolic functions. Their domain restrictions remain part of the formula.

5.13. Logarithmic Differentiation

Taking logarithms before differentiating is useful when products, quotients, or variable exponents make direct differentiation cumbersome.

5.14. Implicit Differentiation

If a relation F(x,y)=0 defines y locally as a function of x, differentiation gives

F_x + F_y y' = 0,

and thus y'=-F_x/F_y when F_y≠0.

5.15. Parametric Differentiation

For x=x(t), y=y(t),

dy/dx = (dy/dt)/(dx/dt)

when dx/dt≠0.

5.16. Higher-order Derivatives

Repeated differentiation gives f'', f''', and f^(n). Higher derivatives describe curvature, acceleration, local approximation, and differential-equation structure.

5.17. Fermat's Theorem and Local Extrema

At an interior local extremum where f is differentiable,

f'(a)=0.

This is a necessary condition, not a sufficient one.

5.18. Rolle's Theorem

If f is continuous on [a,b], differentiable on (a,b), and f(a)=f(b), then some c∈(a,b) satisfies

f'(c)=0.

5.19. Mean Value Theorem

If f is continuous on [a,b] and differentiable on (a,b), some c satisfies

f'(c) = [f(b)-f(a)]/(b-a).

5.20. Generalized Mean Value Theorem

Cauchy's mean value theorem applies the same idea to two functions and provides a foundation for several limit results.

5.21. Taylor Formula

Near a,

f(x)=Σ_(k=0)^n f^(k)(a)(x-a)^k/k! + R_n(x).

The remainder controls the error between the function and its polynomial approximation.

5.22. Indeterminate Forms and L'hôpital's Rule

Under the hypotheses of L'Hôpital's rule, certain 0/0 or ∞/∞ limits can be transformed using derivatives of numerator and denominator. The rule does not apply merely because a quotient looks complicated; the indeterminate form and regularity conditions must first be established.

5.23. Constancy and Monotonicity Tests

On an interval, f'=0 throughout implies constancy under standard differentiability assumptions. The sign of f' determines increasing or decreasing behavior.

5.24. Extremum Tests

Critical points arise where f'=0 or the derivative is undefined. Sign changes of f' or the second-derivative test can classify suitable critical points.

5.25. Marginal Equality in Optimization

Let profit be

Π(x) = R(x) − C(x).

If the functions are differentiable and an optimum lies in the interior of the feasible region, a local-extremum candidate must satisfy

Π'(x) = R'(x) − C'(x) = 0.

Hence the interior candidate condition is

R'(x) = C'(x),

equality of marginal revenue and marginal cost.

This equality is not a proof of optimality. A critical point can be a minimum, maximum, or a stationary non-extremum. First-derivative sign changes, second-derivative information, domain boundaries, and operational constraints must also be examined.

If the feasible set is a closed interval [a,b], the endpoint values must be compared with interior critical points to determine an absolute maximum. With multiple decision variables or constraints, the problem becomes the multivariable and constrained optimization studied in Calculus II.

5.26. Convexity, Concavity, and Inflection Points

The sign of f'' provides a local test for convexity/concavity where the second derivative exists. An inflection point requires a change in concavity; f''=0 alone is not sufficient.

5.27. Procedure for Curve Sketching

A systematic sketch can combine domain, symmetry, intercepts, asymptotes, first-derivative monotonicity, extrema, second-derivative concavity, and limiting behavior.

5.28. Differentials as Local Error Propagation

The differential is not only notation for integration. Around a point x, a small perturbation Δx produces the first-order approximation

Δy ≈ f'(x) Δx.

For several independent input uncertainties this idea becomes the basis of sensitivity analysis. In one variable it already provides a practical engineering check: if a sensor input has bounded error, the derivative estimates how strongly that error is amplified by the model near the operating point.

The approximation is local. When Δx is not small, higher-order Taylor terms can dominate and the linear estimate should not be treated as an exact bound.

5.29. L'hôpital's Rule as a Theorem About Limits

L'Hôpital's rule applies to specific indeterminate forms under differentiability and limit conditions. It is not a generic algebraic simplifier. Before differentiating numerator and denominator, the original limit should be classified and the hypotheses checked.

When direct factorization, a standard limit, or a series expansion gives the result more transparently, those routes are usually easier to verify.

5.30. Local Linearization and Differential Approximation

A derivative is more than the slope of a tangent. It gives the best first-order linear approximation of a smooth function in a small neighborhood. For a small change dx around x,

f(x+dx) ≈ f(x) + f'(x) dx

and the corresponding differential is

dy = f'(x) dx.

This also provides a first-order model for propagating input uncertainty to the output. For sufficiently small changes,

abs(dy) ≈ abs(f'(x)) abs(dx),

so a large abs(f'(x)) indicates high local sensitivity. For relative error, input and output scales must also be considered rather than relying only on absolute change.

In engineering this idea appears in sensor tolerances, calibration, and first-order numerical error analysis. Its multivariable counterpart uses gradients and Jacobians and is developed in Calculus II.

5.31. Taylor Remainder and Approximation Error

A Taylor polynomial approximates a function through local derivative information up to a selected order. The part not represented by the polynomial is the remainder. Under suitable smoothness conditions, the Lagrange form can be written as

R_n(x) = f^(n+1)(xi) (x-a)^(n+1) / (n+1)!,

where xi lies between a and x. The formula exposes the main sources of approximation error: distance from the expansion point, approximation order, and the magnitude of the next derivative.

A higher degree is not automatically a better engineering choice. Coefficient cost, floating-point effects, and the region over which the local model remains meaningful also matter. Taylor expansion is therefore a tool for building an error budget, not merely for manipulating series.

5.32. Necessary and Sufficient Conditions at an Optimum

For a differentiable interior local extremum, the condition

f'(x*) = 0

is generally necessary, but not sufficient. A zero derivative creates a critical-point candidate. Classification still requires a sign test, a second-derivative test, or examination of the problem boundaries.

f''(x*) > 0  -> sufficient local condition for a strict minimum
f''(x*) < 0  -> sufficient local condition for a strict maximum

If f''(x*) = 0, the test is inconclusive and higher-order behavior or derivative signs must be examined. On a closed interval, endpoints must be compared with interior critical points when searching for a global optimum.

The same distinction appears in numerical optimization: reaching a state resembling grad = 0 is not by itself a proof that a desired global optimum has been found.

5.33. Dimensional Analysis and the Engineering Meaning of Derivatives

A mathematically valid expression can still be physically meaningless. Dimensional analysis is one of the cheapest engineering checks: quantities added together must have compatible dimensions, and differentiation or integration changes units in predictable ways.

A derivative is local sensitivity as well as slope. Large derivatives can amplify small input or measurement errors.

An integral represents accumulation: flow to quantity, velocity to position, power to energy. With sampled data, numerical integration also introduces discretization and sampling choices.

5.34. Checking a Result by an Independent Route

Analytical work can be cross-checked numerically: derivatives with finite differences, integrals with numerical quadrature, and limits with sampling plus algebraic transformation. Numerical checks do not replace proof, but they can reveal sign, scale, and transcription errors.

In engineering applications, units are part of the equation. Differentiation and integration change dimensions; dimensionally inconsistent expressions remain physically invalid even if the algebra looks plausible.

The assumptions of a theorem matter. Continuity, differentiability, integrability, and domain restrictions can determine whether a formula applies at all.

5.35. From Derivatives and Optimization to Learning

The most direct connection between Calculus I and AI is the derivative. A learning algorithm often changes model parameters in order to reduce an objective or loss function. The derivative is therefore not merely the slope of a curve; it is local information about how a small parameter change affects the result.

Consider a one-parameter model:

ŷ = w x
L(w) = (y - ŷ)²

The derivative dL/dw indicates the local direction of change. A basic gradient-descent step is:

w_(k+1) = w_k - η dL/dw

If the learning rate η is too large, the iteration may oscillate or diverge; if it is too small, convergence can be slow. This is a direct application of local function behavior.

The chain rule opens the path to backpropagation. For a composition:

y = f(g(x))

the derivative propagates through the composition:

dy/dx = f'(g(x)) g'(x)

Backpropagation applies this idea systematically to a multi-parameter computational graph. Calculus I does not by itself complete the theory of neural-network training, but it explains why reverse propagation of derivatives is possible.

Limits and continuity provide a related foundation. Local optimization arguments depend on assumptions about regularity and differentiability. Not every loss is differentiable everywhere; piecewise-linear functions such as ReLU have points where the classical derivative is undefined and implementations use appropriate derivative conventions or subgradient ideas.

Integration connects through probability. Expectations and marginal probabilities for continuous variables are integrals. Expected risk can be written as:

R(f) = E[L(Y, f(X))]

and is often approximated empirically from samples when the true distribution is unknown.

Calculus I is not a machine-learning course. Its role is to provide the derivative, chain rule, limit, and integral concepts that make optimization and statistical learning mathematically expressible. Multivariable gradients and Hessians are developed in Calculus II, while vector and matrix representation is developed in Linear Algebra.

5.36. Derivative Sensitivity Near Saturation

Performance and network models provide direct systems applications for classical mathematics. In Calculus I, derivatives quantify local sensitivity to small input changes, while integrals quantify accumulated effects of rates, flows, or densities over an interval. These ideas become the foundation for later optimization and differential-equation models of latency, capacity, and dynamics. The goal is not to memorize another application area, but to state clearly which system behavior each mathematical assumption represents.

• Study how the derivative of response time with respect to utilization grows near saturation.
• State the interval in which a local linear approximation is valid.
• Include sensitivity, not only the minimizer, in optimization results.

Bounding the error of a linear approximation. A derivative does more than describe instantaneous change: it can quantify how accurately a costly expression may be approximated. To estimate √104 around x₀ = 100, take f(x) = √x, f(100) = 10, and f'(100) = 1/20. The first-order approximation gives f(104) ≈ 10 + 4/20 = 10.2. Reliability is assessed through the second derivative, f''(x) = −1/(4x^(3/2)). On [100,104], the bound |f''(x)| ≤ 1/4000 yields |R₂| ≤ (1/2)(1/4000)·4² = 0.002 for the Taylor remainder. The actual value is approximately 10.198039, and the observed discrepancy lies within that bound. The computation therefore ends not at the approximation itself, but at a justified error limit. Moving the expansion point farther from the target requires a new remainder estimate.

Unit 6: Indefinite Integrals

Choose an integration method after examining whether the integrand contains composition, products, or rational expressions. For substitution, verify that the differential factor is correctly obtained; for integration by parts, expect a simpler remaining integral. Differentiate an indefinite integral to verify it. For definite integrals, transform the bounds along with the variable and check the result's sign and units in its geometric or physical context.

6.1. Antiderivative

F is an antiderivative of f on an interval if

F'(x)=f(x).

Any two antiderivatives on the same interval differ by a constant.

6.2. Indefinite Integral Concept

∫ f(x) dx = F(x)+C.

The constant C represents the family of antiderivatives.

6.3. Basic Integration Formulas

Standard formulas reverse elementary differentiation rules, such as

∫ x^n dx = x^(n+1)/(n+1)+C,  n≠-1
∫ dx/x = ln|x|+C
∫ e^x dx = e^x+C.

6.4. Substitution

If u=g(x), then

∫ f(g(x))g'(x) dx = ∫ f(u) du.

Substitution is the inverse pattern of the chain rule.

6.5. Integration by Parts

From the product rule,

∫ u dv = uv - ∫ v du.

6.6. Integration of Rational Functions

A proper rational function can often be decomposed into partial fractions after factoring the denominator over the relevant field. Improper rational functions are first divided polynomially.

6.7. Binomial Integrals

Integrals involving powers of x, a+b x^n, or related binomial expressions can sometimes be reduced by substitutions determined by rational-exponent relations. Not every such integral has an elementary antiderivative.

6.8. Trigonometric Integrals

Powers and products of sine, cosine, tangent, secant, and related functions are handled through identities and substitutions chosen from parity and derivative structure.

6.9. Irrational-function Integrals

Expressions containing square roots of quadratic forms can often be simplified through trigonometric or algebraic substitutions. The correct substitution depends on the sign pattern of the quadratic expression.

6.10. Non-elementary Antiderivatives

Some elementary functions do not possess elementary antiderivatives. Examples lead to special functions or numerical integration. The absence of an elementary closed form does not mean the definite integral is meaningless.

Unit 7: Definite Integrals

7.1. Partition of an Interval

A partition

a=x0<x1<...<xn=b

divides [a,b] into subintervals. Refining the partition is the basis of Riemann integration.

7.2. Riemann Sum and Definite Integral

A Riemann sum has the form

Σ f(ξ_i) Δx_i.

If all sufficiently fine partitions lead to one common limit, that limit is the definite integral

∫_a^b f(x) dx.

7.3. Lower and Upper Darboux Sums

Lower sums use infima on each subinterval and upper sums use suprema. A bounded function is Riemann integrable when the gap between upper and lower sums can be made arbitrarily small.

7.4. Necessary Condition for Integrability

A Riemann-integrable function on a closed interval must be bounded. Boundedness alone is not sufficient.

7.5. Integrability Tests

Every continuous function on [a,b] is Riemann integrable. Bounded functions with only sufficiently mild sets of discontinuities are also integrable under broader criteria.

7.6. Properties of the Definite Integral

Linearity, interval additivity, order comparison, and absolute-value bounds are fundamental properties.

7.7. Mean Value Theorem for Integrals

If f is continuous on [a,b], some c∈[a,b] satisfies

∫_a^b f(x)dx = f(c)(b-a).

7.8. Fundamental Theorem of Calculus

If

F(x)=∫_a^x f(t)dt

with continuous f, then F'(x)=f(x). Conversely, if F'=f, then

∫_a^b f(x)dx = F(b)-F(a).

This connects accumulation and differentiation.

7.9. Substitution in a Definite Integral

A change of variables transforms both integrand and limits. Updating the limits avoids mixing variables in the final expression.

7.10. Integration by Parts in a Definite Integral

∫_a^b u dv = [uv]_a^b - ∫_a^b v du.

7.11. Area of Planar Regions

The signed integral measures net area relative to the axis. Geometric area requires nonnegative accumulation, or subtraction of lower functions from upper functions over the relevant intervals.

7.12. Recovering a Total Quantity from a Marginal Quantity

When a derivative represents the local rate of change of a total quantity, a definite integral accumulates those changes and recovers the total change. If marginal cost C'(x) is known, then

C(x) − C(x₀) = ∫[x₀,x] C'(u) du,

so that

C(x) = C(x₀) + ∫[x₀,x] C'(u) du.

The term C(x₀) is an initial condition. The arbitrary constant of an indefinite integral becomes a physically meaningful baseline once an initial value is known. If production is measured from zero, C(0) can represent fixed cost.

The same structure appears throughout engineering:

velocity    → ∫ velocity dt    = displacement
current     → ∫ current dt     = charge
power       → ∫ power dt       = energy
flow rate   → ∫ flow rate dt   = accumulated quantity

A definite integral is therefore not merely geometric area; it is the accumulated effect of a rate. Dimensional analysis confirms the interpretation because the units of the integrand are multiplied by the units of the integration variable.

7.13. Consumer and Producer Surplus

Two area-based applications of the definite integral are consumer surplus and producer surplus. Let quantity be q, market price be p₀, and equilibrium quantity at that price be q₀.

If the inverse demand curve is p = D(q), the total willingness to pay for quantities from 0 to q₀ is modeled by the area under the curve. The actual expenditure is p₀q₀, so consumer surplus is

CS = ∫[0,q₀] D(q) dq − p₀q₀.

If the inverse supply curve is p = S(q), the minimum aggregate amount represented by the supply curve is

∫[0,q₀] S(q) dq.

Subtracting this from market revenue gives producer surplus:

PS = p₀q₀ − ∫[0,q₀] S(q) dq.

Geometrically these are areas between curves and a horizontal market-price line. The interpretation requires the demand and supply functions to be meaningful on the chosen interval, the equilibrium condition to be defined correctly, and the model assumptions to be stated. The integral calculation alone does not validate the economic model.

7.14. Arc Length of a Curve

For y=f(x) with appropriate smoothness,

L = ∫_a^b sqrt(1+[f'(x)]^2) dx.

7.15. Volume of Solids of Revolution

Disk/washer and shell methods express volume as definite integrals. The formula depends on the axis of revolution and chosen slicing geometry.

7.16. Area of Surfaces of Revolution

For suitable rotation about an axis, surface area involves the arc-length factor multiplied by the radius to the axis, for example

S = 2π ∫ y sqrt(1+(y')^2) dx

when the geometric assumptions fit.

7.17. Center of Mass and Moment of Inertia

Integrals weighted by density give mass, first moments, and moments of inertia. The correct integrand follows from physical geometry and the distance to the selected axis.

7.18. Approximate Evaluation of Definite Integrals

When no convenient antiderivative exists, numerical quadrature such as midpoint, trapezoidal, and Simpson-type rules approximates the integral from sampled function values. Their errors depend on step size and smoothness.

7.19. Definite Integrals as Accumulated Quantities

The definite integral can represent more than geometric area. If q(t) is a rate, then

∫ q(t) dt

is the accumulated quantity over the interval. Current integrated over time gives charge; velocity integrated over time gives displacement; power integrated over time gives energy.

This interpretation is useful when checking dimensions. If the integrand has units U/time, the integral over time must have units U. Dimensional inconsistency often exposes an incorrect substitution or missing scale factor before numerical evaluation.

Comparing an analytic integral with numerical estimates. Let velocity over 0 ≤ t ≤ 2 s be v(t) = 3t² m/s. Since velocity is nonnegative, distance equals displacement: s = ∫[0..2] 3t² dt = [t³]₀² = 8 m. With spacing h = 1 s, the composite trapezoidal rule uses v(0) = 0, v(1) = 3, and v(2) = 12, giving s_T = (h/2)[v(0)+2v(1)+v(2)] = 9 m. Simpson's one-third rule instead gives s_S = (h/3)[v(0)+4v(1)+v(2)] = 8 m; under its standard conditions it integrates polynomials through degree three exactly. Method selection depends not just on convenient formulas but on sample spacing, signal smoothness, and the desired error bound. Here the analytic result provides an independent reference for numerical error.

7.20. Error Bounds and Partition Choice in Definite Integration

A numerical approximation to a definite integral requires an error estimate as well as a partition count. If \(f\) has a continuous second derivative and \(h=(b-a)/n\), the composite trapezoidal-rule error satisfies

\[ \lvert E_T\rvert\leq\frac{(b-a)h^2}{12}\max_{x\in[a,b]}\lvert f''(x)\rvert. \]

Without a bound on the second derivative, this expression does not supply a numerical guarantee. Under the stated smoothness conditions, halving the step size can reduce the leading truncation error by roughly a factor of four; roundoff and integrand behavior remain relevant.

For \(f(x)=x^2\) on \([0,1]\), the exact integral is \(1/3\). One trapezoid gives \(T_1=1/2\), whereas two equal subintervals give \(T_2=3/8\). Their respective errors, \(1/6\) and \(1/24\), differ by a factor of four and agree with the second-derivative error expression. The example illustrates convergence but does not establish the same rate for every function.

Uniform partitions may be inefficient near sharp variation or singularities. Adaptive subdivision allocates intervals using local error estimates. Agreement between two successive approximations alone is not an independent guarantee that the true error is small.

Unit 8: Improper Integrals

8.1. Types of Improper Integral

An integral is improper when the interval is unbounded or the integrand becomes unbounded at an endpoint or an interior point. It is defined through limits of proper integrals.

8.2. Convergence and Divergence

An improper integral converges only if the defining limit exists and is finite. Otherwise it diverges. Splitting at every singular point is necessary.

8.3. Fundamental Formulas for the First Type

For an infinite endpoint,

∫_a^∞ f(x)dx = lim_(b→∞) ∫_a^b f(x)dx.

An integral over (-∞,∞) is defined by separate one-sided improper integrals; cancellation of two divergent sides is not ordinary convergence.

8.4. Convergence Tests for the First Type

Comparison and limit comparison are useful for nonnegative integrands. A standard reference is

∫_1^∞ dx/x^p

which converges exactly when p>1.

8.5. Fundamental Formulas for the Second Type

If f is unbounded near a,

∫_a^b f(x)dx = lim_(c→a+) ∫_c^b f(x)dx.

Interior singularities require splitting the interval and checking both sides separately.

8.6. Convergence Tests for the Second Type

The comparison model

∫_0^1 dx/x^p

converges exactly when p<1. Comparison and asymptotic equivalence often decide more complicated cases.

Unit 9: Polar Coordinates and Parametric Geometry

Cartesian coordinates are not the most natural representation for every curve. Polar coordinates simplify geometry defined by angle and radial distance, while parametric representations are natural when a curve is traced by an independent parameter. This unit does not introduce a separate calculus; it transfers the derivative and integral concepts already developed in the course to other coordinate descriptions.

9.1. Conversion Between Cartesian and Polar Coordinates

Represent a point by (r, θ), where r is signed radial distance from the origin and θ is an angle from the chosen reference axis.

The basic conversion relations are

x = r cos θ
y = r sin θ
r^2 = x^2 + y^2.

When r ≠ 0, writing only tan θ = y/x loses quadrant information. Computational systems therefore commonly use a two-argument function such as atan2(y,x).

Polar representation is not unique. For example,

(r, θ)
(-r, θ+π)
(r, θ+2kπ)

can describe the same geometric point. This nonuniqueness matters in curve tracing and in selecting integration limits.

9.2. Reading a Polar Curve

A polar curve has the form

r = f(θ).

Before plotting many sample points, inspect its structure:

1. determine the period of f(θ),
2. find angles for which r=0,
3. inspect extrema of |r|,
4. test useful symmetries,
5. remember that a negative r places the point in the opposite direction.

Common sufficient symmetry tests are

replacing θ by -θ leaves the equation unchanged      -> x-axis symmetry
replacing θ by π-θ leaves it unchanged               -> y-axis symmetry
replacing θ by θ+π leaves it unchanged               -> origin symmetry

Failure of one such test does not by itself prove that the geometric curve has no corresponding symmetry.

9.3. Tangent Slope of a Polar Curve

A polar curve is a parametric Cartesian curve:

x(θ) = r(θ) cos θ
y(θ) = r(θ) sin θ.

Therefore

dx/dθ = r'(θ) cos θ - r(θ) sin θ
dy/dθ = r'(θ) sin θ + r(θ) cos θ,

and when dx/dθ ≠ 0,

dy/dx =
[r'(θ) sin θ + r(θ) cos θ]
/
[r'(θ) cos θ - r(θ) sin θ].

A zero numerator with nonzero denominator gives a horizontal-tangent candidate; a zero denominator with nonzero numerator gives a vertical-tangent candidate. When both vanish, the quotient formula is inconclusive and the local curve must be analyzed more carefully.

9.4. Area in Polar Coordinates

For a small angular increment, a sector with nearly constant radius has area

dA ≈ (1/2) r^2 dθ.

Hence the area swept by r=f(θ) is

A = (1/2) ∫[α,β] r(θ)^2 dθ.

For a region between outer radius R(θ) and inner radius r(θ),

A = (1/2) ∫[α,β] [R(θ)^2 - r(θ)^2] dθ.

The difficult part is usually not the formula but the angular interval. If the chosen interval traces the same region more than once, the integral counts that region more than once. Symmetry arguments must therefore be accompanied by a check of how many times the curve covers the geometric region.

9.5. Arc Length of a Polar Curve

Substituting x=r cos θ and y=r sin θ into the parametric arc-length formula gives

L = ∫[α,β] √[r(θ)^2 + r'(θ)^2] dθ.

The two terms have different geometric roles: r dθ represents angular motion and dr radial motion. Using only the angular contribution is incorrect when the radius changes.

9.6. Surface Area Generated by a Polar Curve

A polar curve can be treated parametrically. If (x(t),y(t)) is revolved about the x-axis over a suitable interval, the surface area is

Sx = 2π ∫ y(t) √[x'(t)^2 + y'(t)^2] dt

when the radius is nonnegative.

With the polar parametrization

x(θ) = r cos θ
y(θ) = r sin θ
ds = √(r^2 + r'^2) dθ,

the geometric-radius form is

Sx = 2π ∫ |r(θ) sin θ| √[r(θ)^2 + r'(θ)^2] dθ.

The absolute value emphasizes that a radius of revolution is a nonnegative geometric distance. For revolution about the y-axis, the radius is |x| = |r cos θ|.

9.7. Parametric Representation as the Unifying Language

Cartesian and polar curves become instances of the same parametric idea:

Cartesian:  r(t) = (t, f(t))
Polar:      r(θ) = (f(θ) cos θ, f(θ) sin θ)
General:    r(t) = (x(t), y(t)).

This removes the need to memorize unrelated formulas. Arc length, for example, is fundamentally

L = ∫ ||r'(t)|| dt,

with each coordinate system supplying its own expression for r'(t).

Limits, derivatives, and integrals of vector-valued functions are developed more systematically in Calculus II.

9.8. Ownership of Neighboring Topics

Sequences, infinite series, power series, and Taylor series belong to Calculus II in this course map. Newton-Raphson is an iterative root-finding algorithm and is therefore kept in Numerical Analysis rather than repeated here. This keeps Calculus I focused on one-variable continuous analysis and its geometric applications.

9.9. Validation of Derivatives and Areas of Parametric Curves

For a parametric curve (x(t), y(t)), the tangent slope is (dy/dt)/(dx/dt) when dx/dt is nonzero. At dx/dt = 0, the quotient is undefined and one must distinguish vertical tangents, stationary points, and higher-order behavior. A geometric point reached at several parameter values may represent different local branches.

For a closed curve, ∮x dy gives signed area. Reversing orientation changes its sign. The parameter interval and orientation must therefore be established before interpreting the integral as geometric area.

References

• Ahmet Yesevi Üniversitesi Bilgisayar Mühendisliği Bölümü. Matematik I (TBIL101) ders materyalleri.

• Albert-László Barabási. Network Science.

• Ian Goodfellow, Yoshua Bengio, Aaron Courville. Deep Learning. MIT Press, 2016.

• Joel Hass, Christopher Heil, Maurice D. Weir, Przemyslaw Bogacki. Thomas' Calculus, 15th Edition. Pearson, 2022. (Pearson)

• Michael Spivak. Calculus, 4th Edition. Publish or Perish, 2008.

• Mor Harchol-Balter. Performance Modeling and Design of Computer Systems: Queueing Theory in Action. Cambridge University Press, 2013.

• Petter Holme; Jari Saramäki (eds.). Temporal Network Theory. 2nd ed. Springer Nature Switzerland, 2023.

• Richard L. Burden, J. Douglas Faires, Annette M. Burden. Numerical Analysis, 10th Edition. Cengage, 2016.

• Tom M. Apostol. Calculus, Volume I, 2nd Edition. Wiley, 1967.

• Walter Rudin. Principles of Mathematical Analysis, 3rd Edition. McGraw-Hill, 1976.