“The Physics of Euler’s Formula | Laplace Transform Prelude” — 3Blue1Brown

Episode summary

Chapter 1 of Grant Sanderson’s Laplace transform trilogy — the prerequisite that makes the next two chapters (2026-04-20-3blue1brown-but-what-is-a-laplace-transform and 2026-04-20-3blue1brown-why-laplace-transforms-are-so-useful) tractable. The 28-minute lesson does two things in tandem: (1) re-derives Euler’s formula e^(it) = cos(t) + isin(t) from the defining property of e (it’s its own derivative) by treating the differential equation x’ = ix as a geometric statement that “velocity is always a 90-degree-rotated copy of position,” which forces motion around a unit circle, and (2) motivates why complex exponentials matter for physics by walking through the damped harmonic oscillator (mass on a spring): guess x(t) = e^(st), substitute into the ODE mx” + mux’ + kx = 0, find that s satisfies a quadratic whose roots are inherently complex, and discover that the complex-valued solution e^(iomega*t) — combined with its conjugate via linearity — gives the real-valued cosine solution that physical intuition demanded. The conceptual punchline: complex exponentials e^(st) are the “atoms of calculus.” Linear ODEs always factor into a polynomial in s (fundamental theorem of algebra), whose roots in the s-plane encode the system’s natural modes — left-half-plane = decay, imaginary axis = oscillation, right-half-plane = blow-up. Closes by previewing that real-world driven equations (the optics-prism damped oscillator with an external cosine forcing term) break the dumb e^(st) trick because solutions are constrained combinations of exponentials — and the Laplace transform is the systematic generalization that handles them.

Key arguments / segments

• [0:00] Trilogy roadmap. This chapter sets up the mental frameworks; chapters 2 and 3 develop the Laplace transform proper. Main characters throughout: exponential functions e^(st), with s allowed to take complex values
• [1:02] Why complex exponents at all? The audience splits — some viewers know the e^(πi) = -1 result by heart, others find it baffling. Either way, worth deriving from the defining property of e (e^t is its own derivative) rather than from the Taylor series
• [2:03] Geometric interpretation of x’ = x. Think of x(t) as the position of a point on the number line; the derivative is velocity. x’ = x means velocity vector = position vector at every moment. Initial condition x(0) = 1 → velocity is initially 1 to the right; the farther right you go, the faster you move → exponential growth, no formula needed
• [3:00] Same geometric move for x’ = 2x (velocity is twice position → faster blow-up) and x’ = -0.5x (velocity is a 180-rotated half-length copy of position → exponential decay toward 0). Now the load-bearing extension: what if x’ = ix? Multiplication by i is a 90-degree rotation. Velocity is always perpendicular to position
• [4:04] The only motion satisfying “velocity perpendicular to position with unit magnitude” is rotation around the unit circle. This is what e^(it) means geometrically. After π units of time, you’ve traversed half the circle → e^(iπ) = -1
• [6:01] Sanderson’s clarification of the notation: “When you input a complex value, the expression really has very little to do with repeated multiplication, and honestly not that much to do with the number e.” The literal computation is the Taylor series; the behavior is much more cleanly captured by “this function is its own derivative.” People get used to using e^x as notational shorthand and really think nothing of it. This is the cleanest “stop trying to make the notation literal, learn the behavior” line in the corpus
• [7:03] The S-plane. Each point on the complex plane represents a value of s and thus an entire function e^(st). Pure imaginary s = oscillation; pure real positive s = growth; pure real negative s = decay; mixed = spiral (decaying or growing oscillation)
• [9:00] Engineers’ S-plane vocabulary. Imaginary part of s = how fast it oscillates and which direction. Real part of s = whether magnitude grows or shrinks. The whole equation x’ = s*x is a differential equation. Setup pivot: having seen how a differential equation can teach what complex exponents mean, a different ODE can motivate why you’d care about complex exponents in the first place
• [10:01] The damped harmonic oscillator. Mass on a spring at position x(t). Newton’s second law + Hooke’s law: F = ma → mx” = -kx. Spring stiffness k. Add damping (friction-like): -mux’ velocity-proportional resistance. Final equation: mx” + mux’ + kx = 0
• [12:02] Reminder about ODEs: there’s no single solution. Different (initial position, initial velocity) pairs give distinct solution functions. Solving the equation = finding the family
• [13:02] The “bizarre trick”: guess x(t) = e^(st) for some constant s. Substitute. The chain rule gives x’ = se^(st), x” = s^2e^(st). Factor out e^(st), and what’s left is the requirement that ms^2 + mus + k = 0 — a quadratic in s, mirror image of the original ODE. Sanderson’s editorial: “guessing and checking like this sort of just feels like asking the student to know the answer ahead of time. And also your physical intuition is telling you that an exponential is probably not really how this mass on a spring behaves.” A desire to make this trick systematic and generalizable is what leads to the Laplace transform
• [14:00] Without damping (mu=0), the quadratic gives s = ±sqrt(-k/m) — inherently complex, whether you wanted it or not. Define omega = sqrt(k/m). The two solutions are e^(±iomegat)
• [15:00] Squeeze a real-valued solution out of the complex one. Two paths: (1) ignore the imaginary part (animation-friendly but ad-hoc), or (2) add the two complex solutions e^(iomegat) + e^(-iomegat) — by Euler’s formula, the result is 2cos(omegat), which is real. Path (2) is the one that generalizes
• [17:00] Why you can add solutions: the equation is linear — sum of solutions is a solution, scaling preserves solution-hood, scaled-and-summed combinations are solutions. Scaling coefficients can themselves be complex — they affect amplitude and initial phase
• [18:01] Sensible objection: if the goal was to get cos(omegat), why not just guess cos(omegat) directly? Sanderson’s concession: “Isn’t this just way more sensible? Why complicate things with complex numbers?” — and his answer is that cosine-and-sine guessing doesn’t generalize
• [19:03] The generalization payoff: turn on damping. With mu ≠ 0, the quadratic formula gives s with both an imaginary AND negative-real part. Plug into e^(st) → both decays AND oscillates. The cosine guess wouldn’t have anticipated this. Increase mu enough and the imaginary parts vanish — solutions become pure decay → spring is overdamped
• [21:02] Generalization to nth-order linear ODEs. Equation = sum of constant-coefficient derivative terms set to zero. Substitute e^(st), each derivative picks up a factor of s, factor out e^(st), result is a degree-n polynomial in s. Fundamental theorem of algebra: polynomials always factor into linear terms over the complex numbers, exposing n roots. Each root is a valid s. Linearity → family of solutions = arbitrary complex-coefficient combinations of these n exponentials. The S-plane root distribution IS the system’s mode structure
• [23:03] Where the trick breaks: real-world driven equations. The damped harmonic oscillator with an external cosine forcing term (came up in the optics-prism video — modeling charges in glass driven by an incoming light wave at a frequency unrelated to the natural resonance). Here the family of solutions is no longer freely-tunable scaled exponentials; the coefficients are constrained. The dumb e^(st) trick doesn’t work directly
• [24:01] But it brings you closer than you’d expect. Driven solutions still look like a combination of four specific exponentials — just with constrained coefficients. Understanding exactly how big those coefficients are as a function of the driving frequency IS the prism phenomenon. “You can think about e^(st) as the atoms of calculus.” Complicated functions describing the world break into these parts — as long as s is allowed to be complex
• [25:02] The key question the next two chapters answer: given an unknown function and a differential equation, how do you actually find the exponential decomposition? How do you know which s-values, with which coefficients, build up a particular solution? Answer: the Laplace transform. Connection to Fourier flagged — Laplace generalizes Fourier to a much larger family of functions
• [26:04 closing] Preview of chapters 2-3: when you use a Laplace transform, the differential equation turns into algebra for essentially the same reason the dumb trick worked — taking a derivative of e^(st) is the same as multiplication by s. The Laplace transform translates functions into a new language where e^(st) are the fundamental units, and in that language differentiation looks like multiplication

Notable claims

• e^(st) are the “atoms of calculus.” Sanderson’s load-bearing metaphor — complicated functions in physics decompose into combinations of complex exponentials, and this is the operational reason the Laplace transform is so universally useful for ODEs
• The geometric interpretation of x’ = i*x as “velocity perpendicular to position with unit magnitude” forces circular motion — and is a more conceptually durable derivation of Euler’s formula than the Taylor-series substitution. The Taylor series is the computation; the differential-equation-and-rotation is the behavior
• “When you input a complex value, the expression e^x really has very little to do with repeated multiplication, and honestly not that much to do with the number e.” Sanderson’s clarification of the notational misdirection. The pedagogical point: focus on what an expression DOES, not on the literal arithmetic of its symbols
• For an nth-order linear ODE with constant coefficients, the substitution x = e^(st) reduces solution-finding to root-finding of a degree-n polynomial in s. The fundamental theorem of algebra guarantees n roots over the complex numbers, and each root is a natural mode of the system
• The S-plane is the “configuration space of natural modes”: real part of s tells you decay/growth, imaginary part tells you oscillation frequency. Reading dynamics off pole positions is faster than computing time-domain solutions explicitly. This frame is the load-bearing intuition that the next two chapters formalize
• Linear ODE solutions form a vector space — closed under addition and complex scalar multiplication. The two-exponential basis (for a quadratic) or n-exponential basis (for an nth-order) is the eigenbasis of the differential operator
• Driven (inhomogeneous) ODEs break the freely-tunable-coefficient property but preserve the exponential-decomposition property — solutions are still combinations of e^(st) terms, just with constrained coefficients. This is the gap the Laplace transform is designed to close
• The bizarre dumb-trick of “guess x = e^(st)” is not a hack; it’s the discrete-spectrum version of what the Laplace transform does in the continuous-spectrum case. The trick “works because differentiation of e^(st) becomes multiplication by s” is the same fact underneath both methods

Why this is in the vault

This is the prerequisite-setup chapter for the Laplace-transform pair already filed at 2026-04-20-3blue1brown-but-what-is-a-laplace-transform and 2026-04-20-3blue1brown-why-laplace-transforms-are-so-useful — together they form a 3-video trilogy that maps cleanly onto the change-of-coordinates-as-core-problem-solving-move thesis that CA-023 already canonicalizes. This chapter contributes the ground-floor justification: why the exponential basis is the natural coordinate system for linear ODEs in the first place. The “atoms of calculus” framing is the most editorially useful way to explain to a non-specialist audience why so many tools across math, signal processing, and ML reduce to “decompose into complex exponentials and operate in the new basis.” Second reason: the geometric derivation of Euler’s formula via x’ = i*x is the most durable pedagogical version of the formula in the corpus — Sanderson’s “velocity perpendicular to position with unit magnitude → motion on the unit circle” is more intuitive and more transferable than any algebraic Taylor-series argument. Third: the “this notation has very little to do with what the symbols literally suggest” aside is a vault-grade reminder for anyone teaching or writing about technical concepts — the meaning of an expression is what it DOES, not what it LOOKS LIKE.

Mapping against Ray Data Co

• Coordinate-change-as-core-move concept (CA-023) — completes the Laplace trilogy on the source list. This chapter is the third of three chapters in the Laplace series. Chapters 2 and 3 are already cited as CA-023 sources 5 and 6; this chapter is the prerequisite. Adding it brings CA-023 to 7 sources total. Honest editorial assessment: this chapter strengthens CA-023 less than the chapter-2/chapter-3 pair did, because the prerequisite role means it sets up the change-of-basis move without explicitly performing it. But it adds two things: (a) the “atoms of calculus” metaphor as a memorable rhetorical anchor for change-of-basis arguments, and (b) the “x’ = i*x → motion on a circle” derivation as the most durable in-vault explanation of why the exponential basis is “natural” for linear systems. Recommended action: add as 7th source in CA-023 entry, but don’t change canon-tier status. Mapping strength: medium / supporting source rather than canonical exemplar
• “Atoms of calculus” — newsletter-worthy frame. The metaphor “complex exponentials are the atoms of calculus, the way prime numbers are the atoms of integers” is a remarkably clean rhetorical device for explaining why so many ML and signal-processing tools converge on Fourier/Laplace/spectral methods. Newsletter angle: “What Are the Atoms of Your Data?” — a piece on representation choice as the load-bearing decision in any ML pipeline. Mapping strength: medium / newsletter-hook
• “The notation has little to do with what it literally suggests” — vault-discipline reminder. Sanderson’s casual dismissal of the “e^x must mean repeated multiplication” intuition is exactly the kind of meta-comment that should be applied across the vault when reviewing technical-pedagogy writing. The exact same trap exists for: matrix multiplication (looks like vector-by-vector summation, IS function composition), softmax (looks like normalized exponentiation, IS argmax-with-temperature), attention scores (look like dot products, ARE relevance-weighted retrievals). Add to RDCO writing rubric: for any technical operator, write what it DOES before writing what it LOOKS LIKE. Mapping strength: medium / writing-rubric
• Geometric ODE intuition — exemplar for “skip the algebra, draw the vector field.” Sanderson’s x’ = i*x → unit circle move generalizes: any first-order ODE can be visualized as “at every point, the velocity vector is some function of the position vector.” For RDCO data-engineering writing on dynamic systems (rate limiters, retry queues, autoscaling), this is the right pedagogical move — show the phase-portrait, not the Laplace solution. Mapping strength: medium / pedagogical-template
• Linearity-and-superposition as the precondition for the trick. Sanderson is explicit that the dumb trick of substituting e^(st) only works because the equation is linear. Lesson for system design: when you can structure your domain to be linear (additive errors, independent failures, decomposable workloads), you unlock the “decompose into modes, solve each mode, recombine” pattern. When you can’t (driven oscillator with cosine forcing, neural net with nonlinear activation), you need richer machinery (Laplace transform, deep-learning training). This is a clean architectural-discipline lesson. Mapping strength: medium / architectural-heuristic
• The “right method, wrong scale” failure mode (cross-link to Tao part 1). This chapter’s analog of Aristarchus’ Sun-distance miscalibration: the cosine-vs-exponential debate at [18:01]. Cosine guessing is correct method, restricted scope — works for undamped oscillator, fails for damped oscillator, fails completely for driven. Exponential guessing is the same method generalized. The lesson — prefer the more general method even when the special-case alternative is cheaper, because generality is what survives the next problem — maps directly to RDCO skill-design. Mapping strength: light / skill-design heuristic

Related

• 2026-04-20-3blue1brown-but-what-is-a-laplace-transform — chapter 2 of this trilogy; this chapter motivates the e^(st) substitution, chapter 2 builds the integral that systematically extracts the s-values
• 2026-04-20-3blue1brown-why-laplace-transforms-are-so-useful — chapter 3 of this trilogy; applies the transform to the driven damped oscillator that this chapter flags as “the dumb trick won’t reach this”
• 2026-04-20-3blue1brown-tao-cosmic-distance-ladder — Tao’s “look at y to measure x” indirect-inference move is the methodological cousin of Sanderson’s “guess e^(st) and let the algebra reveal s” — both are “let the structure tell you what to ask”
• 2026-04-20-3blue1brown-five-puzzles-thinking-outside-the-box — canon-tier source for CA-023; this chapter’s e^(st)-as-natural-coordinate argument is the differential-equations instance of the meta-thesis
• concepts/CANDIDATES — CA-023 entry; this chapter adds as 7th source