Calculus I and II: A Practical Guide to Limits, Derivatives, Integrals, and Series

A limit describes the value a function approaches as its input approaches a specified point. The function need not be defined at that point. For example, (x² − 1)/(x − 1) is undefined at x = 1, but factoring gives x + 1 whenever x ≠ 1. Consequently, its limit as x approaches 1 is 2. This distinction is essential: nearby behavior and the value at the point are separate questions. Continuity at x = a requires that f(a) exist, that the limit exist, and that the two agree. Formally, a finite limit L means that for every tolerance ε > 0, some δ > 0 ensures |f(x) − L| < ε whenever 0 < |x − a| < δ.

Limits and continuity matter when a model contains thresholds, singularities, or piecewise rules. Consider f(x) = |x|/x for x ≠ 0. Approaching zero from the right gives 1; approaching from the left gives −1. Because those one-sided limits disagree, the two-sided limit does not exist. A graph may suggest this result, but the formulas establish it. In computational work, testing progressively smaller inputs can provide evidence, yet floating-point rounding and cancellation can make numerical tables misleading. Simplify analytically first whenever possible, and use numerical checks as confirmation rather than proof.

Differential calculus begins with the derivative f′(x) = lim as h approaches 0 of [f(x + h) − f(x)]/h. For f(x) = x², the quotient simplifies to 2x + h, giving f′(x) = 2x. Geometrically, this is the tangent slope; physically, it is an instantaneous rate. If position is s(t) = 3t² + 2t meters with t measured in seconds, velocity is s′(t) = 6t + 2 meters per second. Differentiating again gives acceleration of 6 meters per second squared. Tracking units makes the interpretation testable and often exposes modeling errors.

Derivative rules turn this definition into an efficient calculation system. The product rule gives (uv)′ = u′v + uv′, while the chain rule gives the derivative of f(g(x)) as f′(g(x))g′(x). For y = x² exp(−3x), combining both rules yields y′ = exp(−3x)(2x − 3x²). Omitting the inner derivative −3 is a common mistake. Implicit differentiation handles relationships that are inconvenient to solve explicitly: differentiating x² + y² = 25 gives 2x + 2y y′ = 0, so y′ = −x/y where y ≠ 0. At points where y = 0, that finite-slope formula does not apply.

Optimization starts with a model and its feasible domain. Suppose a rectangular enclosure uses 100 meters of fencing along three sides, with a wall forming the fourth. If x is each perpendicular side, the parallel side is 100 − 2x, and area is A(x) = 100x − 2x² for 0 ≤ x ≤ 50. The derivative A′(x) = 100 − 4x vanishes at x = 25. Since A″(x) = −4, the area is concave downward throughout the domain. Comparing with the zero-area endpoints confirms the global maximum: dimensions of 25 by 50 meters produce 1,250 square meters.

Practitioners also use derivatives for sensitivity analysis. The linear approximation f(a + Δx) ≈ f(a) + f′(a)Δx predicts how a small input change affects an output. For the area of a circle, A = πr², this gives ΔA ≈ 2πrΔr, or ΔA/A ≈ 2Δr/r. Thus, a small relative radius error produces approximately twice that relative area error. This is a local approximation, not an exact uncertainty guarantee. Reliable calculus for engineering requires checking whether the perturbation is small enough and whether omitted physical effects matter more than the mathematical approximation.

Integral calculus formalizes accumulation through limits of sums. Partition an interval into small pieces, multiply a representative function value by each piece’s width, and add the contributions. Under suitable conditions, these Riemann sums approach the definite integral. If q(t) = 2t + 1 liters per minute is a flow rate over 0 ≤ t ≤ 3 minutes, the accumulated volume is the integral of q(t), equal to [t² + t] from 0 to 3, or 12 liters. For signed velocity, however, an integral gives displacement; total distance requires integrating the absolute value of velocity.

The Fundamental Theorem of Calculus connects differentiation and integration. If f is continuous and F(x) is the integral of f(t) from a to x, then F′(x) = f(x). Also, if G′ = f on the interval, the definite integral from a to b equals G(b) − G(a). This turns an accumulation problem into endpoint evaluation. For example, the integral of 3x² from 1 to 2 is [x³] from 1 to 2 = 7. Indefinite integrals instead describe families of antiderivatives, so the corresponding answer is x³ + C. The constant disappears only when endpoint subtraction is performed.

Useful integration techniques reverse familiar derivative structures. Substitution reverses the chain rule: in the integral of 2x cos(x²), set u = x² and du = 2x dx, obtaining sin(x²) + C. Integration by parts reverses the product rule and states that the integral of u dv equals uv minus the integral of v du. To integrate x exp(x), choose u = x and dv = exp(x) dx. The result is x exp(x) − exp(x) + C. Differentiate this expression to verify it. For definite substitution problems, either transform the bounds into the new variable or return to the original variable before evaluation.

Other integrals require algebraic decomposition, trigonometric identities, or limits at problematic boundaries. For example, the integral of x^(−p) from 1 to infinity is defined as the limit of the integral from 1 to R as R grows without bound. It converges to 1/(p − 1) when p > 1 and diverges when p ≤ 1. A function approaching zero therefore does not guarantee a finite accumulated total: 1/x approaches zero, yet its integral diverges logarithmically. Before applying any numerical method to an unbounded interval or singular integrand, establish convergence and identify the difficult region.

An infinite series converges when its sequence of partial sums approaches a finite value. The geometric series 1 + r + r² + … equals 1/(1 − r) for |r| < 1. Outside that range it diverges. More general infinite series require tools such as comparison, ratio, integral, or alternating-series tests. A necessary condition is that the terms approach zero, but it is not sufficient: the harmonic series Σ 1/n diverges. Power series introduce an interval of convergence, and endpoints must be checked separately. This prevents using a valid-looking expansion where its convergence assumptions fail.

Taylor series use derivatives at a center a to construct terms f⁽ⁿ⁾(a)(x − a)ⁿ/n!. For exp(x) around zero, the cubic polynomial is 1 + x + x²/2 + x³/6. At x = 0.1, it gives approximately 1.1051667. Taylor’s theorem bounds the remainder by M|x|⁴/24 when the fourth derivative is bounded in magnitude by M between zero and x. Here M = exp(0.1), so the error is less than 4.61 × 10⁻⁶. This bound makes the approximation actionable. Smoothness alone does not guarantee that a function equals its infinite Taylor series; controlling the remainder is the essential justification.

Numerical integration is useful when measurements define the integrand or an elementary antiderivative is unavailable. The composite trapezoidal rule uses equally spaced samples with step h = (b − a)/n and computes h[f(a)/2 + Σ f(a + ih) + f(b)/2], where the sum runs from i = 1 to n − 1. For x² on [0, 1] with n = 4, it returns 0.34375, compared with the exact value 1/3. With n = 8, it returns 0.3359375. The error drops by a factor of four, consistent with second-order convergence for sufficiently smooth functions. In Python, NumPy can represent sample arrays, while SciPy provides adaptive quadrature tools.

Numerical differentiation demands similar care. The central difference [f(x + h) − f(x − h)]/(2h) has truncation error of order h² for sufficiently smooth f, but making h extremely small can amplify rounding and measurement noise. A sound workflow checks domains and units, derives an analytical reference where available, and repeats computations at finer resolutions. Compare symbolic derivatives with finite differences at several ordinary points, and compare numerical integrals with known special cases. Report tolerances and convergence behavior rather than every displayed digit. In data science, automatic differentiation often supplies computational derivatives, while calculus remains necessary to interpret them.