📖 Core Concepts
Core Concepts: Calculus II: Integration Techniques, Series, and Differential Equations
1. What This Subject Is About
Calculus I gave you two operations, the derivative (instantaneous rate of change) and the integral (accumulated total), and the Fundamental Theorem that joins them. You learned that differentiation is mechanical. Any function built from elementary pieces can be differentiated by following rules: product rule, chain rule, quotient rule. Integration is not like that. It is the inverse operation, and inverses are harder. Multiplying 37 × 41 is routine, but factoring 1517 takes ingenuity. Calculus II begins by facing this asymmetry.
The course has three big themes. You will see them repeatedly in physical chemistry, kinetics, spectroscopy, and pharmacology.
- Undoing differentiation when the answer is not obvious. Integration techniques (parts, trigonometric substitution, partial fractions) are systematic ways of running the differentiation rules backward.
- Taming infinity. Improper integrals add up infinitely long regions. Series add up infinitely many numbers. Both ask the same question: when does an infinite process produce a finite answer? The answers are often surprising.
- Equations whose unknown is a function. In algebra you solve for a number x. In a differential equation you solve for an entire function y(t), given only a rule for how it changes. Nearly every law of nature is written this way: radioactive decay, first-order reaction kinetics, drug elimination, capacitor charging.
A fourth, smaller theme rounds out the course: describing curves in new coordinate systems (parametric and polar). Some shapes, like a circle, a spiral, or the path of a rotating bond, are awkward as y = f(x) but natural in other languages.
One idea ties these together: representation and approximation. A Taylor series represents a complicated function as an "infinite polynomial." Partial fractions represent a complicated rational function as a sum of simple ones. A differential equation represents a whole time-course by a local rule. Much of science consists of choosing the representation in which a hard problem becomes easy. [Stewart, Clegg & Watson, 2020]
A useful mental model: differentiation takes a function apart, and integration tries to put it back together without the instruction manual. Each technique in this course is a page recovered from that manual.
2. Integration by Parts: The Product Rule Run Backward
Every integration technique comes from a differentiation rule. Substitution (from Calc I) is the chain rule backward. Integration by parts is the product rule backward. [Stewart, Clegg & Watson, 2020, §7.1]
Start from the product rule and integrate both sides:
d/dx [u·v] = u·v' + v·u' ∫ u·v' dx = u·v − ∫ v·u' dx Compact form: ∫ u dv = uv − ∫ v du
What this formula actually does: it does not solve the integral. It trades one integral for another. You win when the new integral ∫ v du is simpler than the original. The skill lies in choosing u (the part you will differentiate) and dv (the part you will integrate) so the trade is favorable.
Heuristic for choosing u, "LIATE": pick u as whichever factor comes first in the list:
- Logarithmic
- Inverse trigonometric
- Algebraic
- Trigonometric
- Exponential
The logic is that functions early in the list become simpler when differentiated (ln x → 1/x; x² → 2x → 2). Functions late in the list stay about as complex when integrated (eˣ → eˣ).
Example 1, a polynomial times an exponential:
∫ x eˣ dx u = x, dv = eˣ dx → du = dx, v = eˣ
= x eˣ − ∫ eˣ dx = x eˣ − eˣ + C = eˣ(x − 1) + C
Check by differentiating: d/dx[eˣ(x−1)] = eˣ(x−1) + eˣ = x eˣ. ✓ Always check. Differentiation is cheap and reliable.
Example 2, the "invisible dv" trick: ∫ ln x dx seems to have only one factor. Write it as ∫ ln x · 1 dx with u = ln x and dv = dx:
∫ ln x dx = x ln x − ∫ x · (1/x) dx = x ln x − x + C
Example 3, the looping integral: for ∫ eˣ sin x dx, applying parts twice returns the original integral with a minus sign:
I = eˣ sin x − eˣ cos x − I ⟹ 2I = eˣ(sin x − cos x) I = ½ eˣ(sin x − cos x) + C
The integral seemed to go in a circle, yet the circle turned into an algebraic equation for I. Treating the unknown integral as a symbol you can solve for is a recurring move in mathematics.
Where you'll see it again: integration by parts derives the moments of probability distributions (mean and variance of an exponential lifetime). It also underlies the quantum-mechanical fact that the momentum operator is Hermitian. The "boundary term" uv becomes important in physics.
3. Trigonometric Integrals and Trigonometric Substitution
Two related tools exploit one fact: the Pythagorean identities let you convert between trigonometric functions. [Stewart, Clegg & Watson, 2020, §7.2–7.3]
sin²θ + cos²θ = 1 1 + tan²θ = sec²θ Half-angle (power-reduction): sin²θ = ½(1 − cos 2θ) cos²θ = ½(1 + cos 2θ)
Trigonometric integrals
For ∫ sinᵐx cosⁿx dx, the strategy depends on parity:
- If one power is odd: peel off one factor, convert the rest with sin² + cos² = 1, and substitute. Example: ∫ sin³x dx = ∫ (1 − cos²x) sin x dx. Let u = cos x, so −du = sin x dx.
- If both powers are even: use the half-angle formulas to reduce the powers.
Trigonometric substitution
This is the clever reverse move. When you see a square root of a quadratic, you introduce a trig function to make the root disappear. Each expression matches a right triangle:
Expression Substitution Identity used √(a² − x²) x = a sin θ 1 − sin²θ = cos²θ √(a² + x²) x = a tan θ 1 + tan²θ = sec²θ √(x² − a²) x = a sec θ sec²θ − 1 = tan²θ
Why this is beautiful: consider ∫₀¹ √(1 − x²) dx. Geometrically, this is the area of a quarter of the unit circle, so the answer must be π/4. With x = sin θ the integral becomes ∫₀^{π/2} cos²θ dθ = π/4. The substitution works because it is the geometry of the circle. Trig substitution translates algebra back into the triangle and circle geometry it came from.
Practical warning: after integrating in θ, you must convert back to x. Draw the reference triangle. If x = a sin θ, the opposite side is x, the hypotenuse is a, and the adjacent side is √(a² − x²). Read off whatever trig function you need.
4. Partial Fraction Decomposition: Un-Adding Fractions
In grade school you learned to combine fractions over a common denominator. Partial fractions is the reverse: it takes a complicated rational function and splits it into simple pieces, each of which integrates to a logarithm or an arctangent. [Stewart, Clegg & Watson, 2020, §7.4]
Combining: 1/x − 1/(x+1) = 1/[x(x+1)]
Decomposing: 1/[x(x+1)] = A/x + B/(x+1)
1 = A(x+1) + Bx
x = 0 → A = 1
x = −1 → B = −1
∫ dx/[x(x+1)] = ln|x| − ln|x+1| + C = ln|x/(x+1)| + C
The general recipe for P(x)/Q(x):
- If deg P ≥ deg Q, do polynomial long division first. Decomposition requires a proper fraction.
- Factor Q(x) completely into linear and irreducible quadratic factors.
- Write one term per factor:
- Distinct linear (x − a) → A/(x − a)
- Repeated linear (x − a)ᵏ → A₁/(x−a) + A₂/(x−a)² + … + Aₖ/(x−a)ᵏ
- Irreducible quadratic (x² + bx + c) → (Ax + B)/(x² + bx + c)
- Solve for the constants, either by plugging in convenient x values (roots) or by matching coefficients.
Why it matters chemically: partial fractions appear directly in kinetics. For a second-order reaction A + B → products with unequal initial concentrations a and b, the rate law dx/dt = k(a − x)(b − x) separates into an integral that requires exactly this decomposition:
1/[(a − x)(b − x)] = 1/(b − a) · [ 1/(a − x) − 1/(b − x) ]
Each piece integrates to a logarithm, which is why second-order integrated rate laws contain ln of a concentration ratio.
5. Improper Integrals: Finite Area from Infinite Regions
A definite integral ∫ₐᵇ f(x) dx assumes a finite interval and a bounded function. An improper integral breaks one of these assumptions. The interval may be infinite, or the function may blow up somewhere. We handle it by replacing the infinity with a limit. [Stewart, Clegg & Watson, 2020, §7.8]
Type 1 (infinite interval):
∫ₐ^∞ f(x) dx = lim_{t→∞} ∫ₐᵗ f(x) dx
Type 2 (infinite discontinuity at b):
∫ₐᵇ f(x) dx = lim_{t→b⁻} ∫ₐᵗ f(x) dx
If the limit exists and is finite, the integral converges. Otherwise it diverges.
The benchmark family, p-integrals. Memorize this, because almost every comparison argument reduces to it:
∫₁^∞ 1/xᵖ dx converges if p > 1 (value = 1/(p − 1))
diverges if p ≤ 1
The counterintuitive heart of the matter: 1/x and 1/x² both go to zero as x → ∞, and both look like gentle curves approaching the axis. Yet:
∫₁^∞ 1/x dx = lim ln t = ∞ (diverges) ∫₁^∞ 1/x² dx = lim (1 − 1/t) = 1 (converges)
Going to zero is not enough. The function must go to zero fast enough. The boundary between "fast enough" and "too slow" sits at exactly p = 1. This idea of a sharp threshold will come back with series.
A real paradox, Gabriel's Horn. Rotate y = 1/x (for x ≥ 1) around the x-axis. The volume is π∫₁^∞ x⁻² dx = π, which is finite. The surface area involves ∫₁^∞ (1/x)√(1 + 1/x⁴) dx ≥ ∫₁^∞ 1/x dx = ∞, which is infinite. You could fill the horn with a finite amount of paint, yet never paint its inner surface. The paradox dissolves once you notice that "paint thickness" stops making sense as the horn narrows below any physical thickness. The mathematics itself is consistent.
Comparison test for integrals: if 0 ≤ f(x) ≤ g(x), then:
- ∫ g converges ⟹ ∫ f converges.
- ∫ f diverges ⟹ ∫ g diverges.
This lets you decide convergence without computing anything. For example, e^{−x²} ≤ e^{−x} for x ≥ 1, so ∫₁^∞ e^{−x²} dx converges, even though e^{−x²} has no elementary antiderivative. (Its full value over ℝ is √π, the Gaussian integral that normalizes the bell curve and the Maxwell–Boltzmann distribution.)
6. Sequences and Series: Adding Infinitely Many Numbers
A sequence is an ordered list a₁, a₂, a₃, … . A series is what you get when you try to add them: Σ aₙ. The sum of infinitely many numbers is defined through partial sums. [Stewart, Clegg & Watson, 2020, §11.1–11.2]
Partial sum: Sₙ = a₁ + a₂ + … + aₙ
Σ_{n=1}^∞ aₙ = lim_{n→∞} Sₙ (if this limit exists, the series converges)
The definition already contains a subtlety. A series is a sequence of partial sums in disguise. "Does the series converge?" means "does the sequence S₁, S₂, S₃, … approach a limit?"
The most important series: geometric
Σ_{n=0}^∞ a rⁿ = a + ar + ar² + … = a / (1 − r) if |r| < 1
diverges if |r| ≥ 1
Derivation: Sₙ = a + ar + … + arⁿ⁻¹
rSₙ = ar + … + arⁿ⁻¹ + arⁿ
Sₙ − rSₙ = a − arⁿ ⟹ Sₙ = a(1 − rⁿ)/(1 − r)
As n → ∞, rⁿ → 0 when |r| < 1, which gives a/(1 − r). Zeno's paradox (½ + ¼ + ⅛ + … = 1) is just this formula with a = r = ½.
The most important cautionary series: harmonic
1 + 1/2 + 1/3 + 1/4 + 1/5 + … = ∞
The terms shrink to zero, but the sum grows without bound. The classical proof, due to Nicole Oresme in the 14th century, groups the terms:
1 + ½ + (⅓ + ¼) + (⅕ + ⅙ + ⅐ + ⅛) + … ≥ 1 + ½ + (¼ + ¼) + (⅛ + ⅛ + ⅛ + ⅛) + … = 1 + ½ + ½ + ½ + … = ∞
The growth is very slow. You need about 12,000 terms to pass 10, and about 1.5 × 10⁴³ terms to pass 100. It still diverges. This is the series version of ∫ 1/x dx = ∞.
The nth-term (divergence) test: if lim aₙ ≠ 0, the series diverges. The converse is false, and the harmonic series is the counterexample. Terms going to zero is necessary but not sufficient.
7. Convergence Tests: A Diagnostic Toolkit
For most series you cannot compute the sum exactly. You can often determine whether it converges. Each test below exploits a different structural feature. [Stewart, Clegg & Watson, 2020, §11.3–11.7]
(a) Integral Test: series ↔ integrals
If f is positive, continuous, and decreasing with f(n) = aₙ, then Σ aₙ and ∫₁^∞ f(x) dx either both converge or both diverge. Picture rectangles of width 1 under and over the curve. This gives the p-series result immediately:
Σ 1/nᵖ converges if p > 1, diverges if p ≤ 1
(Euler's famous result: Σ 1/n² = π²/6. A circle constant appears in a sum of reciprocal squares.)
(b) Comparison and Limit Comparison Tests
Compare to a series you already understand, usually geometric or p-series.
Direct: 0 ≤ aₙ ≤ bₙ. Σbₙ converges ⟹ Σaₙ converges.
Σaₙ diverges ⟹ Σbₙ diverges.
Limit: if lim aₙ/bₙ = c with 0 < c < ∞, both behave the same.
The limit version formalizes "only the dominant behavior matters." Σ (3n + 1)/(n³ − 2) behaves like Σ 3/n², so it converges.
(c) Ratio Test: is it secretly geometric?
L = lim_{n→∞} |aₙ₊₁ / aₙ|
L < 1 → converges absolutely
L > 1 → diverges
L = 1 → TEST IS INCONCLUSIVE
The intuition is that if each term is eventually about L times the previous one, the series behaves like a geometric series with ratio L. The ratio test is ideal for factorials and exponentials. For Σ xⁿ/n!, the ratio |x|/(n+1) → 0, so the series converges for every x.
(d) Root Test
L = lim_{n→∞} |aₙ|^{1/n} (same conclusions as the ratio test)
Use it when the terms contain an nth power, such as ((2n+1)/(3n))ⁿ.
(e) Alternating Series Test
For Σ (−1)ⁿ⁻¹ bₙ with bₙ > 0: if bₙ is decreasing and bₙ → 0, the series converges. The partial sums zig-zag, overshooting and undershooting by ever-smaller amounts and closing in on the limit.
1 − 1/2 + 1/3 − 1/4 + … = ln 2 (converges!) Error bound: |S − Sₙ| ≤ bₙ₊₁ (error ≤ first omitted term)
Absolute vs. conditional convergence
- Absolutely convergent: Σ|aₙ| converges. Such series behave like finite sums. You can rearrange them freely.
- Conditionally convergent: Σaₙ converges but Σ|aₙ| does not. The alternating harmonic series is the standard example.
One of the most surprising facts in this course is the Riemann Rearrangement Theorem. A conditionally convergent series can be reordered to sum to any real number you choose, or to diverge. Addition, which you have trusted to be commutative since childhood, stops being commutative for infinite sums unless convergence is absolute. A conditionally convergent series converges only through a delicate cancellation between an infinite positive part and an infinite negative part. Reordering changes which infinity "wins" at each stage.
Strategy summary: first try the divergence test. Then ask what the series "looks like." Geometric or p-series → compare. Factorials or exponentials → ratio. nth powers → root. Alternating signs → alternating series test, then check absolute convergence. Easy to integrate → integral test.
8. Power Series and Radius of Convergence
A power series is a series in which each term contains a power of a variable:
Σ_{n=0}^∞ cₙ (x − a)ⁿ = c₀ + c₁(x − a) + c₂(x − a)² + …
Think of it as a polynomial of infinite degree. It is no longer a single number. It is a function of x, defined wherever the series converges. [Stewart, Clegg & Watson, 2020, §11.8–11.9]
The key theorem: for every power series there is a radius R (with 0 ≤ R ≤ ∞) such that:
- the series converges absolutely for |x − a| < R;
- it diverges for |x − a| > R;
- at the endpoints |x − a| = R, anything can happen, so each endpoint must be checked separately.
R is usually found with the ratio test. The set of x where the series converges is the interval of convergence.
The parent power series is the geometric series viewed as a function:
1/(1 − x) = 1 + x + x² + x³ + … for |x| < 1 (R = 1)
From this one formula, substitution, differentiation, and integration (all valid term-by-term inside the radius) generate many others:
Replace x by −x²: 1/(1 + x²) = 1 − x² + x⁴ − x⁶ + … Integrate: arctan x = x − x³/3 + x⁵/5 − … At x = 1: π/4 = 1 − 1/3 + 1/5 − 1/7 + … (Leibniz series)
A deep puzzle worth thinking about: 1/(1 + x²) is perfectly smooth for every real x. It never blows up. So why does its power series have radius of convergence exactly 1? The answer lies off the real line. In the complex numbers, 1 + x² = 0 at x = ±i, which is a distance 1 from the origin. The radius of convergence equals the distance to the nearest singularity in the complex plane. Real calculus is quietly governed by complex numbers. You will see this again in quantum mechanics and Fourier analysis.
9. Taylor and Maclaurin Series; Error Bounds
Suppose a function has a power series representation. What must the coefficients be? Differentiate the series n times and set x = a. Every term vanishes except the nth, which gives cₙ = f⁽ⁿ⁾(a)/n!. [Stewart, Clegg & Watson, 2020, §11.10–11.11]
Taylor series about x = a:
f(x) = Σ_{n=0}^∞ f⁽ⁿ⁾(a)/n! · (x − a)ⁿ
= f(a) + f'(a)(x−a) + f''(a)/2! (x−a)² + f'''(a)/3! (x−a)³ + …
Maclaurin series: the special case a = 0.
Interpretation: a function's values and derivatives at a single point can, for well-behaved functions, determine the function everywhere within the radius of convergence. The first-order truncation is the tangent line from Calc I. The second-order truncation adds curvature. Each additional term makes the approximation better.
The essential Maclaurin series (memorize these)
eˣ = 1 + x + x²/2! + x³/3! + … R = ∞ sin x = x − x³/3! + x⁵/5! − x⁷/7! + … R = ∞ cos x = 1 − x²/2! + x⁴/4! − x⁶/6! + … R = ∞ ln(1 + x) = x − x²/2 + x³/3 − x⁴/4 + … R = 1 1/(1 − x) = 1 + x + x² + x³ + … R = 1 (1 + x)ᵏ = 1 + kx + k(k−1)/2! x² + … R = 1 (binomial series)
A moment of real beauty, Euler's formula. Substitute x = iθ into the series for eˣ and group real and imaginary parts. The even powers reproduce cos θ and the odd powers reproduce sin θ:
e^{iθ} = cos θ + i sin θ and at θ = π: e^{iπ} + 1 = 0
Exponential growth and circular oscillation turn out to be the same function viewed along different directions in the complex plane. This identity will run through physical chemistry, where wavefunctions are written as e^{ikx} and oscillating solutions of differential equations become complex exponentials.
How good is the approximation? Error bounds
A truncated Taylor series is the Taylor polynomial Tₙ(x). The error is the remainder Rₙ(x) = f(x) − Tₙ(x). Taylor's Inequality bounds it:
If |f⁽ⁿ⁺¹⁾(x)| ≤ M for |x − a| ≤ d, then |Rₙ(x)| ≤ M · |x − a|ⁿ⁺¹ / (n + 1)! for |x − a| ≤ d
The factorial in the denominator grows faster than any power. That is why the series for eˣ, sin x, and cos x converge for every x.
Worked example: approximate sin(0.1) using sin x ≈ x − x³/6.
- The approximation gives 0.1 − 0.000167 = 0.099833.
- The series is alternating, and the next term is x⁵/120 = 8.3 × 10⁻⁸.
- By the alternating series estimate, the error is at most about 10⁻⁷.
This is why physicists confidently write sin θ ≈ θ for small angles.
Caution: a function can have a Taylor series that converges yet does not equal the function. The classic example is f(x) = e^{−1/x²} (with f(0) = 0). All its derivatives at 0 are zero, so its Maclaurin series is identically 0, but f is not the zero function. Functions that do equal their Taylor series are called analytic. Fortunately, nearly all functions you meet in chemistry are analytic.
10. Introduction to Differential Equations
A differential equation relates an unknown function to its derivatives. The solution is a function (really a family of functions), and an initial condition picks out one member of the family. [Stewart, Clegg & Watson, 2020, Ch. 9]
Differential equations are how science writes its laws. We rarely know directly "how much drug is in the blood at time t." We do know a local rule: "the rate of elimination is proportional to the amount present." The differential equation encodes the rule, and solving it recovers the global behavior.
(a) Separable equations
If dy/dx = g(x)h(y), put all the y's on one side and all the x's on the other, then integrate:
dy/dx = g(x) h(y) ⟹ ∫ dy / h(y) = ∫ g(x) dx
The prototype, exponential growth and decay:
dy/dt = k y
∫ dy/y = ∫ k dt ⟹ ln|y| = kt + C ⟹ y(t) = y₀ e^{kt}
k > 0: growth k < 0: decay
Half-life (decay, k = −λ): t₁/₂ = ln 2 / λ
This one equation describes radioactive decay, first-order chemical reactions, bacterial growth with unlimited resources, and first-order drug elimination. It is probably the most important differential equation in science. Its signature property is that the half-life does not depend on the starting amount.
The logistic equation adds a carrying capacity K:
dP/dt = rP(1 − P/K)
It is separable, and solving it requires partial fractions (see Section 4). The result is the S-shaped logistic curve, which appears in cell-population growth and in cooperative binding curves.
(b) First-order linear equations: the integrating factor
Many equations are not separable but are linear:
dy/dx + P(x) y = Q(x)
The trick is to multiply by a cleverly chosen function μ(x) that turns the left side into the derivative of a product. (This is the product rule appearing yet again.)
Integrating factor: μ(x) = e^{∫ P(x) dx}
Then: μy' + μPy = (μy)' because μ' = μP
So: (μ y)' = μ Q ⟹ y = (1/μ) [ ∫ μ(x) Q(x) dx + C ]
Why it works: we want a μ whose derivative equals μP, because then μy' + μPy is exactly the product rule expansion of (μy)'. The function whose derivative is itself times P is the exponential of ∫P. The method is not a magic formula. It comes from asking "what would make this a product rule?"
11. Applications: One Equation, Many Disguises
A satisfying discovery in this course is that systems with no physical connection obey the same differential equation. Learn the mathematics once and you understand all of them.
RC circuits
A resistor R and capacitor C in series with a battery of voltage E obey Kirchhoff's voltage law:
R dQ/dt + Q/C = E
Charging (Q(0) = 0): Q(t) = CE (1 − e^{−t/RC})
Discharging: Q(t) = Q₀ e^{−t/RC}
Time constant: τ = RC (after one τ, charging reaches 1 − 1/e ≈ 63%)
Drug elimination kinetics (pharmacokinetics)
In a one-compartment model with first-order elimination rate constant k, an IV bolus gives exponential decay:
dC/dt = −kC ⟹ C(t) = C₀ e^{−kt} t₁/₂ = ln 2 / k
For a constant IV infusion at rate k₀ into volume of distribution V, the equation is first-order linear:
dC/dt = k₀/V − kC ⟹ C(t) = (k₀ / kV)(1 − e^{−kt})
Steady state: C_ss = k₀ / (kV)
Notice: this has exactly the same form as the charging capacitor. Infusion rate plays the role of battery voltage, and 1/k plays the role of the time constant RC. The clinical rule that a drug reaches steady state in "about 4–5 half-lives" follows from the mathematics: after 5 half-lives, 1 − 2⁻⁵ ≈ 97% of C_ss.
Total drug exposure is an improper integral, the area under the curve (AUC):
AUC = ∫₀^∞ C₀ e^{−kt} dt = C₀ / k
Repeated dosing is a geometric series. Give dose D every τ hours. Between doses, a fraction r = e^{−kτ} of the drug remains. Just after the nth dose:
Amount = D(1 + r + r² + … + rⁿ⁻¹) = D (1 − rⁿ)/(1 − r)
As n → ∞: Amount_max = D / (1 − e^{−kτ})
Accumulation factor = 1 / (1 − e^{−kτ})
Differential equations, exponentials, and geometric series combine in one practical calculation. This is the kind of synthesis that makes Calc II worth learning as one subject rather than as separate topics.
12. Parametric Curves and Arc Length
A parametric curve describes a path by giving both coordinates as functions of a third variable (the parameter, often time t):
x = f(t), y = g(t)
This frees you from requiring y to be a function of x. A circle (x = cos t, y = sin t) fails the vertical-line test, but as a parametric curve it is perfectly natural. Parametric form also contains more information than the shape alone: it records direction and speed of travel. [Stewart, Clegg & Watson, 2020, Ch. 10]
Slope: dy/dx = (dy/dt) / (dx/dt) (provided dx/dt ≠ 0) Arc length: L = ∫ₐᵇ √[ (dx/dt)² + (dy/dt)² ] dt
Where arc length comes from: this is the Pythagorean theorem applied to infinitesimal pieces. Over a tiny time interval dt, the curve moves dx horizontally and dy vertically, so the tiny length is ds = √(dx² + dy²). Divide inside by dt² and multiply outside by dt, then add up all the pieces. The integrand is the speed, so arc length is speed integrated over time, which is distance traveled.
Arc-length integrals are often impossible to evaluate in closed form. That limitation is where series and numerical methods take over.
13. Polar Coordinates and Area in Polar Form
Polar coordinates locate a point by its distance r from the origin and its angle θ from the positive x-axis:
x = r cos θ y = r sin θ r² = x² + y² tan θ = y/x
Shapes with rotational character become simple. A circle of radius a is just r = a. Spirals (r = θ), cardioids (r = 1 + cos θ), and rose curves (r = cos 3θ) have short polar equations but messy Cartesian ones. [Stewart, Clegg & Watson, 2020, §10.3–10.4]
Area in polar coordinates:
A = ∫_α^β ½ r² dθ = ½ ∫_α^β [f(θ)]² dθ
Why the ½ and the r²? In Cartesian coordinates, we slice area into thin rectangles. In polar coordinates, we slice it into thin pie wedges. A circular sector with radius r and angle dθ has area ½ r² dθ, which is the fraction dθ/2π of the full circle's area πr². Choose the slicing shape that matches the coordinate system, and the formula follows.
Polar arc length follows from the parametric formula with x = r cos θ and y = r sin θ:
L = ∫_α^β √[ r² + (dr/dθ)² ] dθ
Practical warning: polar curves can trace over themselves. For example, r = cos 2θ traces its four petals once as θ goes from 0 to 2π, but r = cos 3θ traces its three petals completely in just θ ∈ [0, π]. Always sketch the curve and determine the θ-range that traces each region exactly once.
14. Connections to Peptide Chemistry
This is a mathematics course, but nearly every topic reappears when you study peptides and proteins. Here is a preview of where these tools lead.
Kinetics of peptide reactions → separable and linear ODEs
Hydrolysis of a peptide bond in excess water follows pseudo-first-order kinetics. Its rate law is dy/dt = −ky, so peptide concentration decays exponentially with a half-life of ln 2/k. Peptide bonds are thermodynamically unstable toward hydrolysis but kinetically very slow to break uncatalyzed; their half-lives at neutral pH are estimated in years. That gap between thermodynamics and kinetics is why proteases are needed.
Many processes are sequential. Examples include a protecting group being removed and then a side reaction, or a peptide being cleaved into an intermediate fragment and then further degraded. For A → B → C with rate constants k₁ and k₂, the intermediate obeys a first-order linear ODE, dB/dt + k₂B = k₁A₀e^{−k₁t}. The integrating factor gives:
[B](t) = k₁[A]₀ / (k₂ − k₁) · ( e^{−k₁t} − e^{−k₂t} )
The rise-and-fall shape of [B] is the same shape as a drug's plasma concentration after oral absorption.
Peptide drug pharmacokinetics → exponentials, improper integrals, geometric series
Therapeutic peptides such as insulin analogs and GLP-1 agonists are designed largely by manipulating the elimination rate constant k. Unmodified GLP-1 has a half-life of only minutes in vivo. Engineered analogs extend this dramatically, for example through fatty-acid acylation that promotes albumin binding. Every design discussion in that field uses the language of Section 11: t₁/₂ = ln 2/k, AUC = C₀/k, and accumulation factors from geometric series.
Solid-phase peptide synthesis → powers and geometric decay
In Merrifield's solid-phase method [Merrifield, 1963], a peptide of n residues requires about n coupling steps (plus deprotections). If each step proceeds with yield p, the overall yield is roughly pⁿ:
p = 0.99, n = 50 → 0.99⁵⁰ ≈ 0.61 p = 0.99, n = 100 → 0.99¹⁰⁰ ≈ 0.37 p = 0.995, n = 50 → 0.995⁵⁰ ≈ 0.78
A tiny per-step inefficiency compounds exponentially. This is why coupling chemistry is optimized so carefully, and why long proteins are usually made biologically or by ligating shorter fragments. Taking logs shows the structure: ln(yield) = n ln p ≈ −n(1 − p) for p near 1. That approximation uses the first term of the Taylor series for ln(1 + x).
Bond vibrations → Taylor series
Expand the potential energy of a bond (for example, the C=O of an amide) in a Taylor series about its equilibrium length r₀. The first derivative vanishes because r₀ is a minimum, so:
V(r) ≈ V(r₀) + ½ V''(r₀)(r − r₀)² + …
This is the harmonic approximation, the quadratic potential of a spring. It is the reason infrared spectroscopy works. The amide I band (dominated by C=O stretching, around 1600–1700 cm⁻¹) is used to estimate protein secondary structure, and its interpretation starts from this second-order Taylor truncation. The cubic and higher terms produce anharmonicity.
Statistical mechanics → series
The molecular partition function q = Σ e^{−εᵢ/kT} is an infinite series. For a harmonic oscillator with evenly spaced levels (measured from the ground state), it is geometric:
q_vib = Σ_{n=0}^∞ e^{−n hν/kT} = 1 / (1 − e^{−hν/kT})
Later, the Zimm–Bragg theory of the helix–coil transition in polypeptides [Zimm & Bragg, 1959] builds partition functions for whole chains. Those calculations use the same ideas of summing weighted terms and analyzing convergence that you learn here.
Backbone conformation → angles and periodic functions
The peptide backbone's conformation is described by two dihedral angles per residue, φ and ψ. Plotting allowed combinations gives the Ramachandran plot [Ramachandran, Ramakrishnan & Sasisekharan, 1963]. Because these are angular variables, they live naturally in polar-style, periodic coordinates: −180° and +180° are the same conformation. Molecular force fields describe torsional energy with periodic cosine sums such as V(φ) = Σ (Vₙ/2)[1 + cos(nφ − γ)]. These are close relatives of the Fourier series you will meet later, which are the trigonometric analogue of Taylor series.
15. Common Misconceptions to Watch Out For
- "If the terms go to zero, the series converges." False. The harmonic series Σ 1/n diverges even though 1/n → 0. Terms going to zero is necessary for convergence but not sufficient. The test only works in one direction: if the terms do not go to zero, the series diverges.
- "Ratio test L = 1 means the series converges (or diverges)." L = 1 tells you nothing. Both Σ 1/n (divergent) and Σ 1/n² (convergent) give L = 1. Switch to another test.
- "The interval of convergence is just (a − R, a + R)." The endpoints must be checked individually. They can differ. For ln(1+x) = Σ (−1)ⁿ⁺¹xⁿ/n, the series converges at x = 1 (alternating harmonic) but diverges at x = −1 (negative harmonic).
- "Integrating straight across a vertical asymptote is fine." Naively, ∫₋₁¹ 1/x² dx = [−1/x]₋₁¹ = −2. That answer is nonsense, since a positive function cannot have negative area. The integrand blows up at x = 0, so the integral is improper and actually diverges. Always scan for discontinuities inside the interval before applying the Fundamental Theorem.
- "An improper integral converges if the series with the same terms converges, and the values are equal." The integral test says Σ aₙ and ∫ f(x) dx converge or diverge together. Their values are generally different.
- "Order doesn't matter when adding." For conditionally convergent series, it does. Rearranging terms can change the sum (Riemann Rearrangement Theorem). Only absolutely convergent series can be rearranged freely.
- "A Taylor series always equals its function." Not necessarily. It equals the function only within the radius of convergence, and only if the remainder → 0. The function e^{−1/x²} has Maclaurin series 0 but is not zero. Also, a series centered at a can never converge farther from a than the nearest (possibly complex) singularity.
- "dy/dx is a fraction, so separating variables is just algebra." Writing dy/h(y) = g(x)dx is a convenient shorthand. Its real justification is the chain rule (integrating both sides with respect to x and substituting). The shorthand is safe in this course, but know why it works. Also check whether dividing by h(y) discarded a constant solution where h(y) = 0.
- "Use P(x) straight from the equation for the integrating factor." First put the equation in standard form, y' + P(x)y = Q(x), with coefficient 1 on y'. For 2y' + 4xy = x, divide by 2 first, so P = 2x, not 4x. Also, a constant of integration inside μ is unnecessary because it cancels.
- "Forgetting + C (or the absolute value in ln|x|)." The antiderivative of 1/x is ln|x| + C, not ln x. In differential equations, the constant of integration is not decoration. It is the parameter fixed by the initial condition. Dropping it means losing the whole family of solutions.
- "Integration by parts always simplifies things." A bad choice of u makes the problem worse. Choosing u = eˣ and dv = x dx in ∫ x eˣ dx produces ∫ x²eˣ/2 dx, which is harder. If the new integral looks worse, swap your choices.
- "Polar area is always ½∫₀^{2π} r² dθ." Many polar curves trace themselves more than once over [0, 2π], or have inner loops. Find the θ-interval that traces each region exactly once, or you will double-count.
- "Half-life depends on how much you start with." For first-order processes it does not. That independence is what makes first-order kinetics special. For zero-order or second-order kinetics, half-life does depend on initial concentration. Don't carry the first-order intuition into other reaction orders.
References
- [Stewart, Clegg & Watson, 2020] Stewart, J., Clegg, D., & Watson, S. Calculus: Early Transcendentals, 9th ed. Cengage Learning, 2020. Chapter 7 (Techniques of Integration, including Improper Integrals), Chapter 9 (Differential Equations), Chapter 10 (Parametric Equations and Polar Coordinates), Chapter 11 (Sequences, Series, and Power Series).
- [Merrifield, 1963] Merrifield, R. B. "Solid Phase Peptide Synthesis. I. The Synthesis of a Tetrapeptide." Journal of the American Chemical Society 85, 2149–2154 (1963).
- [Ramachandran, Ramakrishnan & Sasisekharan, 1963] Ramachandran, G. N., Ramakrishnan, C., & Sasisekharan, V. "Stereochemistry of polypeptide chain configurations." Journal of Molecular Biology 7, 95–99 (1963).
- [Zimm & Bragg, 1959] Zimm, B. H., & Bragg, J. K. "Theory of the Phase Transition between Helix and Random Coil in Polypeptide Chains." Journal of Chemical Physics 31, 526–535 (1959).
⚙️ Mechanisms & Applications
Mechanisms & Applications
Skipper’s organizing idea: Calculus I taught me to move between a quantity and its rate of change. Calculus II teaches me how to make that movement when the formulas are less cooperative—and how to decide whether an infinite calculation actually defines a finite answer. In chemistry and biology, these skills become a language for accumulated exposure, reaction progress, molecular production, and approximation.
The most important habit is to separate three questions: What mechanism am I assuming? What mathematics follows from it? What experiment could distinguish that mechanism from another? An exact solution to the wrong biological model is still the wrong prediction. The mathematical foundations below follow Stewart, Clegg, and Watson, especially Chapters 7–11. [1]
1. Integration techniques: recognize the derivative hiding inside the integrand
An integration method is not merely a trick for changing symbols. Each method reverses a familiar differentiation rule or rewrites the integrand into pieces whose accumulation we understand.
-
First look for substitution: reverse the chain rule. If an integrand contains both a function and its derivative, package the inner function as a new variable. Carry the differential and, for definite integrals, transform the bounds.
Example: a changing rate with an exponential factor ∫₀ᵀ t e^(−kt²) dt, k > 0 Let u = kt², so du = 2kt dt. ∫₀ᵀ t e^(−kt²) dt = (1/2k) ∫₀^(kT²) e^(−u) du = [1 − e^(−kT²)]/(2k).
The exponent must be dimensionless. If t is measured in hours, then k here has units of h−2, not the h−1 of ordinary first-order elimination.
-
Use integration by parts when a product becomes simpler after differentiating one factor. This reverses the product rule:
d(uv) = u dv + v du Therefore: ∫ u dv = uv − ∫ v du. Example: ∫ t e^(−kt) dt Choose u = t and dv = e^(−kt) dt. Then du = dt and v = −e^(−kt)/k. ∫ t e^(−kt) dt = −t e^(−kt)/k − e^(−kt)/k² + C.
The choice works because differentiating t simplifies it, while integrating the exponential preserves its form. Integration by parts is especially useful for moments: integrals that weight observations by time, position, or another variable. Moments later become central to probability distributions and pharmacokinetic residence times.
-
For trigonometric integrals, use identities to expose a substitution. With an odd power of sine, save one sine factor and convert the remaining even power using sin²x = 1 − cos²x. With only even powers, half-angle identities often help.
∫ sin³x cos²x dx = ∫ (1 − cos²x) cos²x sin x dx Let u = cos x, so du = −sin x dx. Result: −u³/3 + u⁵/5 + C = −cos³x/3 + cos⁵x/5 + C.
-
For certain radicals, substitute a trigonometric expression. The goal is to turn a square root into a trigonometric identity.
For a > 0: √(a² − x²): use x = a sin θ √(a² + x²): use x = a tan θ √(x² − a²): use x = a sec θ. Example: √(a² − a² sin²θ) = a|cos θ|. Choose a suitable interval for θ before removing the absolute value.
This is geometry doing algebraic work: a radical becomes a side length in a right triangle.
-
For rational functions, use partial fractions. First perform polynomial division if the numerator’s degree is at least the denominator’s. Then factor the denominator and solve for coefficients. Repeated linear factors require every power up to the multiplicity; irreducible quadratic factors require linear numerators.
1/[x(x + a)] = (1/a)[1/x − 1/(x + a)], a ≠ 0 Thus: ∫ dx/[x(x + a)] = (1/a) ln|x/(x + a)| + C.
Partial fractions transform one difficult accumulation into a sum of simpler ones. Logarithms appearing in integrated reaction-rate laws often come from exactly this decomposition. [1, Ch. 7]
2. From reaction stoichiometry to an integrated rate law
In General Chemistry I, stoichiometry tracked how much reactant disappeared. A differential equation adds how quickly it disappears. For a well-mixed, constant-volume system with the assumed rate law A + B → product, let x(t) be the concentration of product formed, with initial reactant concentrations a and b.
d x/dt = k(a − x)(b − x), x(0) = 0.
Separate:
dx/[(a − x)(b − x)] = k dt.
For a ≠ b:
1/[(a − x)(b − x)]
= [1/(b − a)] [1/(a − x) − 1/(b − x)].
Integrate and apply x(0) = 0:
[1/(b − a)] ln{ a(b − x) / [b(a − x)] } = kt.
The logarithm’s argument is dimensionless, and both sides have units of inverse concentration. When a = b, the displayed formula cannot be used directly; integrating that case separately gives:
dx/dt = k(a − x)² 1/(a − x) − 1/a = kt.
Application to peptide manufacturing: such models can be useful starting points for interpreting coupling-reagent or substrate consumption. But reaction stoichiometry alone does not establish a rate law. Activation steps, competing hydrolysis, resin swelling, and diffusion within solid supports can make real peptide-coupling kinetics substantially more complicated. The equation is a testable model, not a consequence of writing “A + B.”
3. Improper integrals: an infinite duration need not imply an infinite total
An improper integral is defined through a limit. Never treat infinity as an ordinary upper endpoint and simply “plug it in.” For exponential drug elimination, suppose plasma concentration is C(t) = C0e−kt, with k > 0.
AUC₀→∞ = lim(T→∞) ∫₀ᵀ C₀e^(−kt) dt
= lim(T→∞) (C₀/k)(1 − e^(−kT))
= C₀/k.
Units:
(concentration) × (time), for example mg·h/L.
AUC, the area under the concentration–time curve, measures systemic exposure. The mathematical model never reaches exactly zero concentration, yet its total exposure is finite. That is one of the course’s important surprises: an infinitely long tail can contain only a finite amount of area. [1, §7.8; 2]
The tail’s rate of decrease matters:
∫₁^∞ t^(−p) dt converges exactly when p > 1. By contrast, near a singularity: ∫₀¹ t^(−p) dt converges exactly when p < 1.
These are different problems: decay far away versus blow-up near a finite point. If there are several improper endpoints or interior singularities, test each separately. Divergent pieces cannot be made into an ordinary convergent improper integral by canceling opposite infinities.
Integration by parts also gives the first time moment of a monoexponential concentration curve:
AUMC = ∫₀^∞ t C(t) dt = C₀/k². For a one-compartment intravenous bolus model: mean residence time = AUMC/AUC = 1/k. Half-life = ln(2)/k.
Mean residence time and half-life are not interchangeable: one is a mean of a residence-time distribution under the model, while the other is the time required for a concentration to halve. [2]
4. Series: accumulation by infinitely many small contributions
A sequence lists values. A series asks whether their cumulative sum settles toward a limit. For a series ∑an, the object whose convergence matters is the sequence of partial sums SN = a1 + ⋯ + aN.
Check the terms first. If an does not tend to zero, the series diverges. The converse is false: ∑1/n diverges even though 1/n → 0.
Identify a familiar benchmark. Geometric series converge when the absolute value of the common ratio is less than one. The positive-term series ∑1/np converges exactly when p > 1.
Use comparison or limit comparison for positive terms resembling a known benchmark. Use the integral test when the terms come from a positive, continuous, eventually decreasing function.
Use ratio or root tests for factorials, exponentials, or expressions raised to the n-th power.
For alternating signs, test the magnitudes. If they decrease eventually to zero, the alternating series converges. Then test absolute convergence separately.
Ratio test: L = lim(n→∞) |a_(n+1)/a_n|. Root test: L = lim(n→∞) |a_n|^(1/n), when this limit exists. L < 1: absolute convergence. L > 1: divergence. L = 1: inconclusive—not a verdict of divergence. Alternating-series error: If the magnitudes decrease to zero, |sum − S_N| ≤ magnitude of the first omitted term.
Application: repeated dosing. Suppose each instantaneous dose adds D/V to concentration in a one-compartment model, doses are separated by time τ, and elimination is first order. Immediately after the N-th dose, residual contributions add geometrically:
q = e^(−kτ), 0 < q < 1.
C_peak,N = (D/V)(1 + q + q² + ... + q^(N−1))
= (D/V)(1 − q^N)/(1 − q).
At periodic steady state:
C_peak,ss = (D/V)/(1 − q)
C_trough,ss = q C_peak,ss.
“Steady state” here does not mean constant concentration. It means a repeating peak–trough cycle. This model assumes linear kinetics, fixed parameters, and instantaneous input; oral absorption, nonlinear clearance, and multiple distribution compartments require extensions. [1, Ch. 11; 2]
5. Taylor series: local approximations with a measurable trust region
A Taylor polynomial matches a function’s value and successive derivatives at a chosen point. It is a local surrogate whose reliability must be established, not assumed.
P_n(x) = Σ(j=0 to n) [f^(j)(a)/j!] (x − a)^j. If |f^(n+1)(u)| ≤ M between a and x: |f(x) − P_n(x)| ≤ M|x − a|^(n+1)/(n+1)!.
For a general power series ∑cn(x − a)n, the ratio or root test usually determines a radius of convergence R. Test each endpoint separately. Inside the radius, power series can be differentiated and integrated term by term. Also remember that an infinitely differentiable function need not equal its Taylor series: equality requires the remainder to approach zero. [1, §§11.8–11.11]
Example: short-time elimination. For t ≥ 0:
e^(−kt) = 1 − kt + (kt)²/2! − ... Linear approximation: C(t) ≈ C₀(1 − kt). Since the second derivative of e^(−z) is e^(−z) ≤ 1 for z ≥ 0: |e^(−kt) − (1 − kt)| ≤ (kt)²/2. At kt = 0.1: absolute error in C(t)/C₀ ≤ 0.005.
The controlling quantity is the dimensionless product kt, not time alone. Ten minutes may be “short” for one drug and “long” for another. The linear approximation also becomes negative when kt > 1, demonstrating why a local approximation should not be extrapolated indefinitely.
Application to peptide synthesis: if each of N coupling steps succeeds independently with the same probability p, and failures cannot later become the desired sequence, the idealized full-length fraction is pN. Writing p = 1 − ε gives:
ln(full-length fraction) = N ln(1 − ε) ln(1 − ε) = −ε − ε²/2 − ε³/3 − ..., |ε| < 1. For small ε: p^N ≈ e^(−Nε). Example: 0.99^50 ≈ 0.605.
A 99% success rate sounds excellent until it is multiplied across fifty steps. The neglected correction in the exponent begins with Nε²/2, so many repetitions can amplify even small approximation errors. This is an idealized sequence-fidelity calculation, not an isolated manufacturing yield: cleavage, deprotection, side reactions, and purification introduce additional losses.
6. Differential equations: turn a conservation statement into a prediction
A reliable modeling sequence is: define the state variable → state assumptions → write a balance → solve → apply initial conditions → check units and limits.
For a drug amount A(t) in a constant-volume, well-mixed compartment:
Accumulation = input − output dA/dt = R_in(t) − CL·C C = A/V Therefore: dA/dt + kA = R_in(t), k = CL/V.
Clearance CL has units of volume/time; multiplying by concentration gives amount/time. It is an effective proportionality parameter, not necessarily a literal stream of plasma physically removed from the body. [2]
Separable mechanism: no input.
dA/dt = −kA. For A > 0: dA/A = −k dt. Integrate: ln(A/A₀) = −kt. Therefore: A(t) = A₀e^(−kt). Also retain A(t) = 0 as an equilibrium solution.
Dividing by the state variable can discard equilibrium solutions. Check them before separating.
Integrating-factor mechanism: time-dependent input. A first-order linear equation has the form y′ + p(t)y = q(t). Multiply by a function that turns the left side into one product derivative:
μ(t) = exp(∫p(t) dt) μy′ + μpy = μq d(μy)/dt = μq. For A′ + kA = R_in(t): A(t) = A₀e^(−kt) + ∫₀ᵗ e^[−k(t−s)] R_in(s) ds.
The integral has a physical interpretation: input arriving at time s is discounted by the fraction surviving until time t. Present concentration is a weighted memory of past input.
For constant infusion R₀: A(t) = A₀e^(−kt) + (R₀/k)(1 − e^(−kt)). C_ss = R₀/CL. Starting from zero, after n half-lives: C(t)/C_ss = 1 − 2^(−n).
The same equation appears in an RC circuit: dq/dt + q/(RC) = E/R for constant applied voltage E. The electrical time constant RC corresponds to the pharmacokinetic time constant 1/k. Unrelated physical systems share a solution because they share the same balance structure. [1, Ch. 9]
7. Enzyme kinetics: why exponential decay can fail
Cell Biology introduced enzymes as catalysts with finite binding and processing capacity. Under a Michaelis–Menten substrate-depletion model, with constant active enzyme and negligible reverse reaction and product inhibition:
dS/dt = −V_max S/(K_m + S). Separate: (1 + K_m/S) dS = −V_max dt. With S(0) = S₀: V_max t = (S₀ − S) + K_m ln(S₀/S).
Here Vmax has units of concentration/time and Km has units of concentration. The rate law itself relies on enzyme-kinetic assumptions; the integration does not establish those assumptions.
Low substrate, S ≪ Km: dS/dt ≈ −(Vmax/Km)S. Depletion is approximately exponential.
High substrate, S ≫ Km: dS/dt ≈ −Vmax. Depletion is approximately linear because the enzyme is saturated.
This explains why a fixed half-life need not survive a change in concentration. Saturable drug elimination can show similar behavior, although an actual pharmacokinetic model must account for which measured concentration controls the elimination process. Michaelis and Menten’s 1913 work was foundational; the later Briggs–Haldane treatment supplied the widely used quasi-steady-state framework. [3, 4]
8. Experimental interfaces: what Calculus II helps me interpret
This mathematics course introduces no wet-laboratory technique. Its experimental contribution is quantitative: it explains what measurements mean, how to accumulate them, and how to test candidate models. The following are applications of the mathematics, not laboratory procedures taught by the course.
-
Concentration–time measurements by HPLC or LC–MS. Chromatography separates sample components through different interactions with mobile and stationary phases; a detector records a response as compounds elute. In LC–MS, mass-to-charge information contributes selectivity. Calibrated standards connect response to analyte concentration.
There are two different areas: a chromatographic peak area integrates detector response over retention time for one sample; a pharmacokinetic AUC integrates concentration across samples collected at different post-dose times. They are not the same quantity.
Linear trapezoidal estimate: AUC₀→tN ≈ Σ_i [(C_i + C_(i+1))/2](t_(i+1) − t_i). If a terminal exponential phase is adequately established: estimated remaining AUC = C_last/λ_z, where λ_z is the estimated terminal decay constant.
Sampling gaps, assay bias, and an incorrectly identified terminal phase can dominate the error. Analytical validation addresses properties such as selectivity, accuracy, precision, and reportable range. [2, 5]
-
Spectrophotometric enzyme progress curves. Under Beer–Lambert conditions, absorbance is proportional to concentration: Abs = εℓc. Repeated measurements convert an optical signal into a concentration–time curve. Differentiation estimates rates; fitting an integrated rate law uses the curve directly.
Differentiation amplifies noise because it subtracts nearby measurements and divides by a small time interval. Integrated-model fitting avoids that particular operation, but still requires a suitable model, baseline treatment, and measurement-error assumptions.
-
Testing first-order behavior. For exponential decay, ln[C(t)/Cref] plotted against time is linear with slope −k. Systematic curvature can flag model inadequacy, but it does not uniquely identify its cause: distribution, saturation, changing physiology, or assay limitations can all contribute. Log transformation also changes the error structure, so fitting transformed and untransformed data need not produce equivalent estimates.
9. Parametric and polar geometry: accumulate along a path or across a region
A moving object need not trace a graph y = f(x). Parametric coordinates describe each position component as a function of a shared parameter.
For a planar path (x(t), y(t)): speed = √[(dx/dt)² + (dy/dt)²] arc length: L = ∫ₐᵇ √[x′(t)² + y′(t)²] dt. For a spatial path, include z′(t)² under the square root.
In cell migration or microscopy, path length measures total travel, whereas displacement measures only the straight-line separation of endpoints. A cell can travel extensively and finish near where it began. Measurement resolution matters: extra localization noise can artificially lengthen an observed track.
A simple geometric model for a helical trajectory is:
x(θ) = a cos θ y(θ) = a sin θ z(θ) = bθ Length for θ₁ ≤ θ ≤ θ₂: L = √(a² + b²)(θ₂ − θ₁). Axial rise per revolution = 2πb.
This illustrates why distance along a helix exceeds its axial rise. It is a geometric idealization, not a complete atomic model of a peptide backbone.
In polar coordinates, a thin sector has approximate area ½r²Δθ, giving:
Area = (1/2) ∫_α^β r(θ)² dθ. Between outer and inner boundaries: Area = (1/2) ∫_α^β [r_outer(θ)² − r_inner(θ)²] dθ.
Angles must be in radians, and the bounds must trace the intended region without unintended repetition. Polar area is useful for approximately radial structures in imaging, such as colony boundaries or tissue cross-sections, when a suitable radial representation exists. [1, Ch. 10]
10. Historical context: why these methods changed science
Newton and Leibniz, late seventeenth century: independently developed systematic forms of calculus, building on earlier work. Their great unification linked local change with global accumulation. It made differential laws of motion—and later kinetic laws—into tools for predicting entire trajectories.
Infinite expansions predate the familiar names: the Kerala mathematical tradition developed important trigonometric series before European calculus. In seventeenth-century Europe, Newton made extensive use of infinite series, and Nicholas Mercator published the logarithm series in 1668. Series turned difficult functions into successions of simpler arithmetic operations.
Brook Taylor, 1715, and Colin Maclaurin, 1742: Taylor published the general expansion associated with his name; Maclaurin’s later treatment helped establish the expansion about zero. Neither name should be read as a claim that all component ideas originated with that person. Modern error analysis is what turns such expansions from plausible formal manipulations into controlled approximations.
Cauchy and nineteenth-century analysis: increasingly rigorous treatment of limits and convergence clarified when infinite processes were legitimate. This mattered because a formal rearrangement or termwise operation can fail without the right hypotheses.
Michaelis and Menten, 1913; Briggs and Haldane, 1925: quantitative enzyme kinetics connected laboratory measurements to mechanistic mathematical models. The enduring lesson is not just a famous equation, but the need to state which approximations connect a molecular mechanism to an observed rate law. [3, 4, 6]
Skipper’s final checkpoint: Before trusting any result, I should be able to name the conservation law or approximation behind it, check its units, identify its validity range, and describe what evidence would make me revise it. Calculus II is not simply about obtaining answers—it is about learning when an accumulation, an infinite expansion, or a dynamical prediction deserves belief.
References
- Stewart, J., Clegg, D., and Watson, S. Calculus: Early Transcendentals. 9th ed. Cengage, 2020. Chapters 7–11: integration techniques, applications of integration, differential equations, parametric and polar curves, and infinite sequences and series.
- Rowland, M., and Tozer, T. N. Clinical Pharmacokinetics and Pharmacodynamics: Concepts and Applications. 4th ed. Lippincott Williams & Wilkins, 2011.
- Johnson, K. A., and Goody, R. S. “The Original Michaelis Constant: Translation of the 1913 Michaelis–Menten Paper.” Biochemistry 50 (2011): 8264–8269.
- Briggs, G. E., and Haldane, J. B. S. “A Note on the Kinetics of Enzyme Action.” Biochemical Journal 19 (1925): 338–339.
- International Council for Harmonisation. ICH Q2(R2): Validation of Analytical Procedures. Final version, adopted 1 November 2023.
- Katz, V. J. A History of Mathematics: An Introduction. 3rd ed. Addison-Wesley, 2009.
🪬 Skipper's Reflection
1. What Surprised Me Most About This Material
Going into Calculus II, I expected a toolbox of tedious algebraic tricks to solve integrals that I couldn't crack in MATH101. What I actually found was a profound philosophical exploration of infinity. The most surprising realization for me was that infinity isn't just a boundary we can never reach—it is a tool we can harness to calculate finite, tangible realities.
Stewart, Clegg, and Watson (2020) introduce improper integrals by asking us to calculate the area under a curve that extends infinitely along the x-axis. My biological intuition rebelled at this: how can an infinitely long region have a finite area? Yet, by evaluating limits as the bounds approach infinity, we can definitively prove convergence. It fundamentally shifted my perspective: just because a process never technically stops doesn't mean it doesn't settle into a bounded, finite state.
Equally shocking was the Taylor and Maclaurin series. The idea that a transcendental function—like the exponential curves I see in cell growth—can be perfectly reconstructed by stacking an infinite number of increasingly complex polynomials blew my mind. It felt less like mathematics and more like genetic sequencing: taking a complex organism (a function) and decoding its DNA into a sequence of basic polynomial building blocks.
2. Conceptual Breakthrough Moments
My first major breakthrough came with Integration by Parts. At first, $\int u \, dv = uv - \int v \, du$ felt like a random shuffling of symbols. But then it clicked: it is just the Product Rule from Calc I played in reverse. I realized it's a negotiation. You take an impossible integral, break it into two parts, and "trade" a derivative for an integral to create a new, hopefully easier problem. It’s exactly like a coupled reaction in biochemistry (which I touched on in BIOL101), where you pair a non-spontaneous reaction with a highly spontaneous one to push the whole process forward.
The ultimate "eureka" moment, however, was my introduction to Differential Equations (ODEs). In Calculus I, derivatives were static—we were looking at a snapshot of a curve to find its slope. But ODEs allow us to write rules for a dynamic system and watch it evolve. When studying first-order linear equations, the integrating factor initially seemed like dark magic. Why multiply the whole equation by $e^{\int P(x)dx}$? The breakthrough came when I saw that this forces the left side of the equation to collapse elegantly into the derivative of a single product. It’s mathematical enzyme catalysis—introducing an external factor that suddenly makes an impossible transformation proceed smoothly.
3. Connections to My Work with Peptides
This course completely rewired how I view peptide pharmacokinetics, bioavailability, and clinical data.
In the past, when I looked at the 5-day half-life of Tirzepatide or the rapid systemic clearance of BPC-157, I simply accepted the decay charts provided in clinical literature. Now, I understand the underlying engine driving those curves. Drug elimination is fundamentally a separable first-order differential equation: $\frac{dC}{dt} = -kC$. The rate at which the peptide is cleared from the bloodstream is directly proportional to its current concentration. Integrating this gives the classic exponential decay model $C(t) = C_0 e^{-kt}$ (Stewart et al., 2020). I no longer just see a curve; I see the integrated result of biological clearance rates.
Furthermore, Improper Integrals gave me the mathematical foundation for "Area Under the Curve" (AUC), a crucial metric in drug development that measures total drug exposure over time. To calculate total exposure for a dose of Retatrutide, we are literally calculating the improper integral $\int_{0}^{\infty} C(t) dt$. Understanding that this infinite integral converges explains why a patient's total exposure remains finite, even though theoretically, trace amounts of the peptide decay indefinitely.
Even Partial Fraction Decomposition gave me a conceptual framework for peptide degradation. Just as a complex rational function can be broken down into simpler, easily integrated fractions, a long-chain peptide like TB-500 is enzymatically cleaved into smaller, more manageable amino acid sequences in the body. Analyzing the pieces individually is often the only way to understand the behavior of the whole.
4. Questions This Course Raises for Further Investigation
While MATH102 unlocked first-order ODEs for me, it immediately made me realize the limitations of these simple models.
- Multi-compartment models: A basic first-order ODE assumes the body is one single bucket. But lipophilic peptides or complex molecules like Retatrutide partition into fat, liver, and blood at different rates. How do we model these interacting compartments? I suspect I will need to learn systems of differential equations, which likely requires linear algebra.
- Periodic dosing schedules: Taylor series use polynomials to approximate curves. But what if I want to model a patient taking a GLP-1 agonist periodically, layered over their natural circadian rhythms of insulin release? Is there a way to build functions out of sine and cosine waves instead of polynomials? (A brief look ahead suggests this might be something called Fourier Series).
- Error Bounds in Biology: Taylor's Inequality allows us to calculate the maximum error of an approximation. How often is this utilized in computational biology when approximating the minimum energy states of folding proteins?
5. Confidence Rating
Most Confident:
I feel highly confident in Differential Equations (separable and first-order linear) and their applications to exponential growth/decay. Because I have a strong biological anchor for these concepts (pharmacokinetics, cell culture growth from BIOL101), the math feels intuitive and purposeful. I am also very comfortable with Integration by Parts and Improper Integrals.
Least Confident:
I am struggling with the vast array of Convergence Tests for Series (Ratio, Root, Integral, Comparison, Limit Comparison, Alternating). While I understand the mechanics of each test, looking at a random series and intuitively knowing which test to apply still feels like guessing. I need more practice developing an intuition for the "dominant" terms in series.
Additionally, Polar Coordinates and Parametric Curves felt like a detour. While I can calculate area and arc length in polar form, I don't yet see how these coordinate systems apply to the chemical and biological systems I care about. They remain abstract mathematical exercises for me right now.
🗺️ Visual Aids & Key Takeaways
Integration Technique Decision Tree
╔══════════════════════════════════════════════════════════════════════════════════╗
║ CHOOSING AN INTEGRATION TECHNIQUE ║
║ Stewart §7.1–7.5 — the art of recognizing form ║
╚══════════════════════════════════════════════════════════════════════════════════╝
┌─────────────────────┐
│ START: ∫ f(x) dx │
└──────────┬──────────┘
│
┌──────────────▼──────────────┐
│ Can you spot a u and du? │
│ (simple substitution) │
└──────────────┬──────────────┘
│
┌──────────┴──────────┐
YES NO
│ │
┌────────▼────────┐ │
│ U-SUBSTITUTION │ │
│ u = inner fn │ │
│ ∫f(g(x))g'(x)dx│ │
└─────────────────┘ │
│
┌────────────────────▼──────────────────────┐
│ Does the integrand contain a PRODUCT │
│ of two qualitatively different functions? │
│ (e.g., x·eˣ, x²·sin x, ln x·x³) │
└────────────────────┬──────────────────────┘
│
┌──────────────┴──────────────┐
YES NO
│ │
┌──────────▼──────────┐ │
│ INTEGRATION │ │
│ BY PARTS │ │
│ ∫u dv = uv − ∫v du │ │
│ │ │
│ LIATE priority: │ │
│ Logarithms │ │
│ Inverse trig │ │
│ Algebraic (xⁿ) │ │
│ Trig functions │ │
│ Exponentials │ │
│ ← pick u from left │ │
└─────────────────────┘ │
│
┌───────────────────────────────▼──────────┐
│ Does the integrand contain TRIG POWERS │
│ or √(a²−x²), √(a²+x²), √(x²−a²)? │
└───────────────────────────────┬──────────┘
│
┌──────────────────┴─────────────────┐
YES NO
│ │
┌────────────────┴─────────────────┐ │
│ Trig form? │ Radical form? │ │
└──────┬────────┘──────────┬───────┘ │
│ │ │
┌────────────▼────┐ ┌──────────▼──────────┐ │
│ TRIG INTEGRALS │ │ TRIG SUBSTITUTION │ │
│ │ │ │ │
│ ∫sinᵐx cosⁿx dx │ │ √(a²−x²): x=a sinθ │ │
│ odd power → │ │ √(a²+x²): x=a tanθ │ │
│ save one, │ │ √(x²−a²): x=a secθ │ │
│ convert rest │ │ │ │
│ via identity │ └─────────────────────┘ │
│ │ │
│ even powers → │ │
│ half-angle │ │
│ identities │ │
└─────────────────┘ │
│
┌──────────────────────────────────────────▼──┐
│ Is the integrand a RATIONAL function │
│ P(x)/Q(x), degree P < degree Q? │
└──────────────────────────────────────────┬──┘
│
┌──────────────────┴────┐
YES NO
│ │
┌────────────▼──────────┐ │
│ PARTIAL FRACTION │ ┌───────▼──────────────┐
│ DECOMPOSITION │ │ TRY: tables, CAS, │
│ │ │ completing square, │
│ factor Q(x), then: │ │ or creative combo │
│ linear (ax+b) → A/ │ │ of above methods │
│ ax+b │ └──────────────────────┘
│ repeated linear → │
│ A/(ax+b)ⁿ terms │
│ irreducible quad → │
│ (Ax+B)/(ax²+bx+c) │
└────────────────────────┘
A decision tree walking through the four main integration strategies in Calculus II — substitution, integration by parts, trigonometric methods, and partial fractions — with the key diagnostic question at each branch.
Convergence Tests for Series — When to Use What
╔══════════════════════════════════════════════════════════════════════════════════════╗
║ SERIES CONVERGENCE TESTS: A DIAGNOSTIC MAP ║
║ Stewart §11.2–11.7 — knowing which weapon to reach for ║
╚══════════════════════════════════════════════════════════════════════════════════════╝
Given: Σ aₙ (sum from n=1 to ∞)
│
┌───────────────────────▼────────────────────────────────┐
│ FIRST: Does lim(n→∞) aₙ = 0? │
│ If NOT zero → series DIVERGES immediately │
│ (Divergence Test — necessary but not sufficient) │
└───────────────────────┬────────────────────────────────┘
│ aₙ → 0 ✓, so we continue
│
┌─────────────────────────┼──────────────────────────────────┐
│ │ │
▼ ▼ ▼
aₙ contains aₙ = f(n) for aₙ involves
FACTORIALS or continuous, positive, ALTERNATING
nth POWERS decreasing f(x)? signs (−1)ⁿ?
│ │ │
▼ ▼ ▼
┌──────────────┐ ┌──────────────┐ ┌──────────────────┐
│ RATIO TEST │ │ INTEGRAL TEST│ │ ALTERNATING │
│ │ │ │ │ SERIES TEST │
│ L = lim │ │ Converges ↔ │ │ (Leibniz) │
│ |aₙ₊₁/aₙ| │ │ ∫₁^∞f(x)dx │ │ │
│ │ │ converges │ │ Converges if: │
│ L < 1 → │ │ │ │ 1. |aₙ| decreasing
│ CONVERGES │ │ NOTE: gives │ │ 2. aₙ → 0 │
│ L > 1 → │ │ same answer │ │ │
│ DIVERGES │ │ as series, │ │ Error ≤ |aₙ₊₁| │
│ L = 1 → │ │ not the sum │ │ (beautiful!) │
│ INCONCLUS. │ └──────────────┘ └──────────────────┘
└──────────────┘
│
│ Ratio test inconclusive (L=1)? Try:
│
▼
┌───────────────────────────────────────────────────────────────────────────┐
│ aₙ looks like a p-series or geometric? USE COMPARISON │
│ │
│ DIRECT COMPARISON: 0 ≤ aₙ ≤ bₙ │
│ Σbₙ converges ⇒ Σaₙ converges │
│ Σaₙ diverges ⇒ Σbₙ diverges │
│ │
│ LIMIT COMPARISON: lim(aₙ/bₙ) = c, 0 < c < ∞ │
│ ⇒ Σaₙ and Σbₙ both converge or both diverge │
│ (gentler — works when direct comparison fails to find right bound) │
└───────────────────────────────────────────────────────────────────────────┘
┌──────────────────────────────────────────────────────────────────────────┐
│ CHEAT SHEET: KNOWN SERIES │
│ │
│ Geometric: Σ arⁿ |r| < 1 → converges to a/(1−r) │
│ |r| ≥ 1 → diverges │
│ │
│ p-series: Σ 1/nᵖ p > 1 → converges │
│ p ≤ 1 → diverges (p=1 is harmonic: diverges!) │
│ │
│ Root Test: L = lim ⁿ√|aₙ| — best when aₙ = (f(n))ⁿ │
│ same L interpretation as Ratio Test │
└──────────────────────────────────────────────────────────────────────────┘
A decision map for selecting the right convergence test, organized by the structural features of the general term aₙ — the key insight is that the test you choose depends on what aₙ looks like, not on a fixed recipe.
Taylor Series — Approximating Functions as Infinite Polynomials
╔════════════════════════════════════════════════════════════════════════════════╗
║ TAYLOR & MACLAURIN SERIES: THE CENTRAL IDEA ║
║ Stewart §11.10–11.11 — every smooth function has a polynomial soul ║
╚════════════════════════════════════════════════════════════════════════════════╝
THE CORE QUESTION: Can we write f(x) as an infinite polynomial?
f(x) = c₀ + c₁(x−a) + c₂(x−a)² + c₃(x−a)³ + ...
To find the coefficients, differentiate repeatedly at x = a:
f(a) = c₀ → c₀ = f(a)
f'(a) = c₁ → c₁ = f'(a)
f''(a) = 2c₂ → c₂ = f''(a)/2!
f'''(a)= 6c₃ → c₃ = f'''(a)/3!
GENERAL TERM: cₙ = f⁽ⁿ⁾(a) / n!
┌──────────────────────────────────────────────────────────────────────┐
│ TAYLOR SERIES centered at x = a: │
│ │
│ ∞ │
│ f(x) = Σ f⁽ⁿ⁾(a) · (x−a)ⁿ / n! │
│ n=0 │
│ │
│ When a = 0: MACLAURIN SERIES (most common case) │
└──────────────────────────────────────────────────────────────────────┘
ESSENTIAL MACLAURIN SERIES — memorize these cold:
eˣ = 1 + x + x²/2! + x³/3! + x⁴/4! + ... all x ∈ (−∞, ∞)
(converges everywhere)
sin x = x − x³/3! + x⁵/5! − x⁷/7! + ... all x ∈ (−∞, ∞)
(only ODD powers — because sin is an odd function)
cos x = 1 − x²/2! + x⁴/4! − x⁶/6! + ... all x ∈ (−∞, ∞)
(only EVEN powers — because cos is an even function)
1/(1−x) = 1 + x + x² + x³ + ... |x| < 1 only!
(geometric series in disguise)
ln(1+x) = x − x²/2 + x³/3 − x⁴/4 + ... −1 < x ≤ 1
HOW SUCCESSIVE PARTIAL SUMS CLOSE IN ON f(x):
(illustrated for sin x near x = 0)
sin x exact value at x=1: 0.8415...
Partial sum accuracy:
P₁(x) = x → at x=1: 1.0000 (rough)
│
▼ add next term
P₃(x) = x − x³/6 → at x=1: 0.8333 (much better)
│
▼ add next term
P₅(x) = x − x³/6 + x⁵/120 → at x=1: 0.8417 (very close)
│
▼ add next term
P₇(x) = x − x³/6 + x⁵/120 − x⁷/5040 → at x=1: 0.84147 (excellent)
Each term corrects the overshoot of the previous — like zeroing in.
ERROR BOUND (Taylor's Inequality):
|Rₙ(x)| ≤ M · |x−a|^(n+1) / (n+1)!
where M = max of |f⁽ⁿ⁺¹⁾| on the interval of interest.
This tells you HOW MANY TERMS you need for a desired precision.
RADIUS OF CONVERGENCE via Ratio Test:
R = lim |aₙ / aₙ₊₁| (when limit exists)
The series converges for |x − a| < R
Check endpoints separately — they can go either way.
This diagram builds Taylor series from first principles — showing where the coefficients come from, listing the five essential Maclaurin series to memorize, illustrating how partial sums converge numerically, and giving the error bound formula for practical computation.
ODE Solution Methods — Separable vs. First-Order Linear
╔═════════════════════════════════════════════════════════════════════════════════╗
║ ORDINARY DIFFERENTIAL EQUATIONS: TWO KEY SOLUTION STRATEGIES ║
║ Stewart §9.1–9.5 — connecting rates of change to physical reality ║
╚═════════════════════════════════════════════════════════════════════════════════╝
┌────────────────────────────────────┐ ┌────────────────────────────────────┐
│ SEPARABLE EQUATIONS │ │ FIRST-ORDER LINEAR EQUATIONS │
│ dy/dx = g(x) · h(y) │ │ dy/dx + P(x)y = Q(x) │
├────────────────────────────────────┤ ├────────────────────────────────────┤
│ KEY IDEA: │ │ KEY IDEA: │
│ Algebra separates variables — │ │ Multiply both sides by an │
│ x stuff on one side, │ │ integrating factor μ(x) │
│ y stuff on the other. │ │ that makes the left side │
│ │ │ a perfect derivative. │
│ METHOD: │ │ │
│ 1) Rewrite as: │ │ METHOD: │
│ dy/h(y) = g(x) dx │ │ 1) Compute μ(x) = e^(∫P(x)dx) │
│ │ │ │
│ 2) Integrate both sides: │ │ 2) Multiply through: │
│ ∫dy/h(y) = ∫g(x)dx + C │ │ d/dx[μ(x)·y] = μ(x)·Q(x) │
│ │ │ │
│ 3) Solve for y (if possible) │ │ 3) Integrate both sides: │
│ — sometimes implicit is fine │ │ μ(x)·y = ∫μ(x)Q(x)dx + C │
│ │ │ │
│ EXAMPLE: Exponential growth │ │ 4) Solve for y │
│ dy/dt = ky │ │ │
│ dy/y = k dt │ │ EXAMPLE: RC circuit │
│ ln|y| = kt + C │ │ dI/dt + (R/L)I = V/L │
│ y = Ae^(kt) │ │ P(t) = R/L, Q(t) = V/L │
│ │ │ μ = e^(Rt/L) │
│ APPLICATION GALLERY: │ │ │
│ • population growth/decay │ │ APPLICATION GALLERY: │
│ • radioactive decay │ │ • electric circuits (RL, RC) │
│ • Newton's cooling law │ │ • drug concentration over time │
│ • drug elimination (first order) │ │ • mixing problems │
└────────────────────────────────────┘ └────────────────────────────────────┘
PHYSICAL REALITY CHECK — why these equations appear everywhere:
EXPONENTIAL DECAY MODEL (separable):
Rate of change ∝ current amount
│ │
▼ ▼
dC/dt = −k · C (drug elimination)
dN/dt = −λ · N (radioactive decay)
dT/dt = −k(T − Tenv) (Newton's cooling)
All have the same mathematical skeleton → same solution form
→ this is why math built in CHEM101 now pays dividends here
GENERAL SOLUTION ALWAYS: y(t) = y₀ · e^(−kt)
HALF-LIFE: t₁/₂ = ln(2)/k ≈ 0.693/k
A side-by-side comparison of the two ODE techniques covered in Calculus II — separable equations (split variables, integrate both sides) and first-order linear equations (integrating factor method) — with their diagnostic forms, step-by-step methods, and real applications in chemistry and physics.
Key Takeaways
- Integration by Parts follows the formula ∫u dv = uv − ∫v du, and the LIATE mnemonic (Logarithms, Inverse trig, Algebraic, Trig, Exponential) tells you which factor to call u. The idea is to trade a hard integral for an easier one by differentiating one factor and integrating the other. Sometimes you must apply it twice, and occasionally applying it twice to a cyclic integrand like ∫eˣ sin x dx loops back to the original — which you then solve algebraically for the integral. (Stewart §7.1)
- Trigonometric substitution is the correct technique when the integrand contains √(a²−x²), √(a²+x²), or √(x²−a²) — you substitute x = a sinθ, x = a tanθ, or x = a secθ respectively, converting an algebraic radical into a trigonometric expression that cancels under the square root. The payoff is that Pythagorean identities (sin²θ + cos²θ = 1, etc.) eliminate the radical entirely. You must always convert back to x using a reference triangle at the end. (Stewart §7.3)
- Partial fraction decomposition works on any proper rational function P(x)/Q(x) where degree(P) < degree(Q): factor Q(x) completely, then write the integrand as a sum of simpler fractions (A/(ax+b) for linear factors, (Ax+B)/(ax²+bx+c) for irreducible quadratics) whose constants are determined by clearing denominators and matching coefficients or plugging in strategic values of x. If degree(P) ≥ degree(Q), perform polynomial long division first. (Stewart §7.4)
- Improper integrals are defined as limits: ∫ₐ^∞ f(x) dx = lim_{t→∞} ∫ₐᵗ f(x) dx, and the integral converges if and only if that limit exists as a finite number. A canonical result worth memorizing: ∫₁^∞ 1/xᵖ dx converges if and only if p > 1, giving the value 1/(p−1). This is the continuous analogue of the p-series test and directly motivates it. (Stewart §7.8)
- The Divergence Test is always your first move with a series Σaₙ: if lim_{n→∞} aₙ ≠ 0, the series diverges, period. However, the converse is false — the harmonic series Σ1/n has terms tending to zero yet diverges (proven by the integral test), and this is one of the most important counterintuitive facts in analysis. Never conclude convergence from the divergence test alone. (Stewart §11.2)
- The p-series Σ 1/nᵖ converges if p > 1 and diverges if p ≤ 1, and these benchmark series are the backbone of the comparison tests. To apply Limit Comparison, find a benchmark bₙ that behaves like aₙ for large n, compute lim(aₙ/bₙ), and if the limit is a positive finite number c, then Σaₙ and Σbₙ share the same convergence fate. This is almost always cleaner than direct comparison when the inequality direction is ambiguous. (Stewart §11.3–11.4)
- The Ratio Test — compute L = lim|a_{n+1}/aₙ|; converges if L < 1, diverges if L > 1, inconclusive if L = 1 — is the natural choice whenever aₙ contains factorials (n!) or exponentials (rⁿ), because the n! in the denominator of aₙ₊₁ will cancel most of aₙ, simplifying dramatically. The Root Test (L = lim ⁿ√|aₙ|) is preferred when aₙ = (f(n))ⁿ, since the nth root collapses the exponent. Both tests are equivalent in power but differ in computational convenience. (Stewart §11.6)
- The Alternating Series Test (Leibniz): Σ(−1)ⁿ bₙ converges if (i) bₙ is eventually decreasing and (ii) lim bₙ = 0. The error bound is especially elegant: truncating after N terms introduces an error no larger than the absolute value of the first omitted term, |b_{N+1}|. This makes alternating series practically useful for computation — you know exactly how accurate your partial sum is. (Stewart §11.5)
- A power series Σcₙ(x−a)ⁿ always has a radius of convergence R (which can be 0, ∞, or finite) determined by R = lim|cₙ/c_{n+1}|, and the series converges absolutely for all x with |x−a| < R and diverges for |x−a| > R. The endpoints x = a ± R must be tested individually — they can independently converge or diverge, which is why the interval of convergence can be open, half-open, or closed. (Stewart §11.8)
- The five Maclaurin series to know cold are: eˣ = Σxⁿ/n! (all x), sin x = Σ(−1)ⁿx^{2n+1}/(2n+1)! (all x), cos x = Σ(−1)ⁿx^{2n}/(2n)! (all x), 1/(1−x) = Σxⁿ (|x|<1), and ln(1+x) = Σ(−1)^{n+1}xⁿ/n (−1
You generate new series by substituting, differentiating, or integrating these rather than recomputing from scratch — for example, substituting x→−x² into 1/(1−x) immediately gives the series for 1/(1+x²), which integrates to arctan x. (Stewart §11.10) - Taylor's Inequality gives a rigorous error bound: if |f^{(n+1)}(x)| ≤ M on an interval, then the remainder satisfies |Rₙ(x)| ≤ M|x−a|^{n+1}/(n+1)!. In practice, you use this to determine how many terms of a Taylor series are needed to achieve a specified accuracy (e.g., within 0.001). The (n+1)! in the denominator is why Taylor series errors shrink so fast — factorial growth beats any power of |x−a| for |x−a| bounded away from infinity. (Stewart §11.11)
- Separable ODEs — those writable as dy/dx = g(x)h(y) — are solved by algebraically separating variables (dy/h(y) = g(x)dx) and integrating both sides. The exponential growth/decay equation dy/dt = ky is the canonical example, yielding y = Ae^{kt}. The half-life formula t_{1/2} = ln(2)/k ≈ 0.693/k applies universally to any first-order decay process — radioactive nuclei, drug plasma concentration, or Newton's cooling — because they all share this ODE skeleton. (Stewart §9.3)
- First-order linear ODEs of the form dy/dx + P(x)y = Q(x) are solved by the integrating factor method: multiply through by μ(x) = e^{∫P(x)dx}, which converts the left side into the exact derivative d/dx[μ(x)y], making both sides directly integrable. The reason this works is that μ(x)y' + μ(x)P(x)y equals d/dx[μ(x)y] by the product rule — the integrating factor is specifically chosen to create this coincidence. This technique is the foundation for modeling RC circuits, RL circuits, and drug infusion kinetics. (Stewart §9.5)
- Arc length of a parametric curve x = f(t), y = g(t) from t = α to t = β is L = ∫_α^β √((dx/dt)² + (dy/dt)²) dt, which generalizes the Cartesian formula L = ∫√(1+(dy/dx)²) dx. In polar coordinates, the area enclosed by r = f(θ) from θ = α to β is A = (1/2)∫_α^β [f(θ)]² dθ — the factor of 1/2 arises because polar area is built from circular sector elements (area = ½r²Δθ), not rectangular strips. (Stewart §10.2, §10.4)
References
- James Stewart, Daniel Clegg, and Saleem Watson, Calculus: Early Transcendentals, 9th ed. (Cengage Learning, 2020). Chapters 7 (Integration Techniques), 9 (Differential Equations), 10 (Parametric Equations and Polar Coordinates), and 11 (Infinite Sequences and Series). ISBN 978-1-337-61392-7.