Main Portal UC3M 251-14169
Pattern, table Theorem, definition Derivation Algorithm Exam trap Prerequisite
Advanced Maths — Formula SheetUC3M Aerospace Engineering · Units 1 to 6

Unit 1 — Modelling and classification

Pattern Order, linearity and autonomy in five seconds

  • Order = highest derivative that appears. ODE: one independent variable $t$; PDE: partial derivatives in two or more.
  • Linear of order $n$: $a_n(t)x^{(n)} + \dots + a_1(t)\dot x + a_0(t)x = f(t)$. The unknown and its derivatives appear only to the first power, never multiplied together, never inside $\sin$, $e^{(\cdot)}$, $\sqrt{\ }$. The coefficients may depend on $t$ in any way.
  • Homogeneous if $f \equiv 0$. Autonomous if $t$ does not appear explicitly: $\dot x = f(x)$.
  • Nonlinear red flags: $x\dot x$, $(\dot x)^2$, $x^2$, $\sin x$, $e^{x}$, $1/x$.

Theorem The superposition test

an equation is linear and homogeneous exactly when any combination of two solutions is again a solution.

Write the equation as $L[x] = 0$. It is linear when $L[\alpha x_1 + \beta x_2] = \alpha L[x_1] + \beta L[x_2]$ for all constants $\alpha, \beta$.

$L[x] = \ddot x + p(t)\dot x + q(t)x$
the general homogeneous linear operator of order 2
$L[\alpha x_1 + \beta x_2] = \alpha(\ddot x_1 + p\dot x_1 + qx_1) + \beta(\ddot x_2 + p\dot x_2 + qx_2)$
the derivative of a sum is the sum of derivatives, constants come out
$= \alpha\cdot 0 + \beta\cdot 0 = 0$
each bracket vanishes because $x_1$, $x_2$ are solutions

To prove an equation is not linear, one counterexample is enough: for $\ddot u = (\dot u)^2 + 1$, the square of $\dot u_1 + \dot u_2$ produces the cross term $2\dot u_1\dot u_2$, which does not cancel.

Table The models of the course

ModelEquationSolution
Growth / decay$\dot N = kN$$N = N_0e^{kt}$; half-life $t_{1/2} = \ln 2/\lvert k\rvert$
Logistic$\dot N = rN(1 - N/K)$$N = \dfrac{KN_0}{N_0 + (K - N_0)e^{-rt}}$
Newton cooling$\dot T = -k(T - T_a)$$T = T_a + (T_0 - T_a)e^{-kt}$
Mixing tank$\dot Q = r_{in}c_{in} - r_{out}\,Q/V(t)$linear, integrating factor
Free fall$m\ddot x = -mg$$x = x_0 + v_0t - \tfrac12gt^2$
Pendulum$\ddot\theta + (g/l)\sin\theta = 0$small angles, from rest: $\theta = \theta_0\cos\sqrt{g/l}\,t$

Derivation Newton's law of cooling

In short: shift the unknown by the room temperature, $u = T - T_a$, and the equation becomes pure decay $\dot u = -ku$.
$u = T - T_a \Rightarrow \dot u = \dot T$
$T_a$ is constant, its derivative is zero
$\dot u = -ku \Rightarrow u = u_0e^{-kt}$
the decay law, solved by separation
$T(t) = T_a + (T_0 - T_a)e^{-kt}$
undo the shift, $u_0 = T_0 - T_a$
$k = \dfrac1{t_1}\ln\dfrac{T_0 - T_a}{T_1 - T_a}$
one extra reading $T(t_1) = T_1$ fixes $k$ (time-of-death problems)

Units: $[k] = \mathrm{s^{-1}}$, so $kt$ is dimensionless, as the exponent must be.

Algorithm Building a model

  1. Name the unknown, its unit and the independent variable.
  2. Write the balance: rate of change = rate in − rate out (mass, salt, population) or $m\ddot x = \sum F$ (mechanics).
  3. For a tank write the volume first: $V(t) = V_0 + (r_{in} - r_{out})t$; concentration $= Q/V$.
  4. Classify the equation (order, linear, autonomous) and pick the method from Unit 2 or 3.
  5. Check units term by term and the limits $t \to 0$, $t \to \infty$.

Trap Typical modelling slips

  • Using $V_0$ instead of $V(t)$ when inflow and outflow rates differ.
  • Forgetting that $\dot N = kN$ grows for $k > 0$ and decays for $k < 0$: read the sign from the physics.
  • Linearising $\sin\theta \approx \theta$ outside small angles.

Unit 2 — First-order ODEs

Theorem Picard: existence and uniqueness

if $f$ and $\partial f/\partial x$ are continuous around $(t_0, x_0)$, the IVP $\dot x = f(t,x)$, $x(t_0) = x_0$ has exactly one solution near $t_0$.

  • Hypotheses: $f$ continuous and $\partial f/\partial x$ continuous on a rectangle containing $(t_0, x_0)$; enough is Lipschitz in $x$: $\lvert f(t,x) - f(t,y)\rvert \le L\lvert x - y\rvert$.
  • Conclusion: a unique solution on some interval $(t_0 - h, t_0 + h)$; it depends continuously on $x_0$.
  • Consequence: two solution curves never cross, and a non-equilibrium solution never reaches an equilibrium in finite time.

Trap When Picard fails or stops short

EquationWhat goes wrong
$\dot x = \sqrt{x}$, $x(0) = 0$$\partial f/\partial x = 1/(2\sqrt x)$ blows up at $0$: both $x \equiv 0$ and $x = t^2/4$ solve it (no uniqueness)
$\dot x = x^2$, $x(0) = x_0 > 0$$x = \dfrac{1}{1/x_0 - t}$ blows up at $t = 1/x_0$: the theorem is only local
$\dot x = f(t,x)$ with $f$ discontinuousexistence itself may fail

Uniqueness check in an exam: compute $\partial f/\partial x$ and find where it is not continuous; outside those points the solution is unique.

Pattern Autonomous equations and the phase line

$\dot x = f(x)$. Equilibria: $f(x^*) = 0$. Draw the $x$-axis, mark the zeros of $f$, and put an arrow to the right where $f > 0$ and to the left where $f < 0$.

  • Stable (sink): arrows point in on both sides; $f'(x^*) < 0$.
  • Unstable (source): arrows point out; $f'(x^*) > 0$.
  • Semi-stable: same sign of $f$ on both sides; $f'(x^*) = 0$ and $f$ does not change sign.
  • If $f'(x^*) = 0$, the derivative test is silent: read the sign of $f$ on each side.

Derivation Stability from the linearisation

In short: near $x^*$, $\dot\xi \approx f'(x^*)\xi$ for $\xi = x - x^*$, so small perturbations behave like $e^{f'(x^*)t}$.
$x = x^* + \xi \Rightarrow \dot x = \dot\xi$
$x^*$ is constant
$f(x^* + \xi) = f(x^*) + f'(x^*)\xi + O(\xi^2)$
Taylor expansion of $f$ around $x^*$
$\dot\xi \approx f'(x^*)\xi$
$f(x^*) = 0$ at an equilibrium; drop $O(\xi^2)$ for small $\xi$
$\xi(t) \approx \xi_0e^{f'(x^*)t}$
decays if $f'(x^*) < 0$, grows if $f'(x^*) > 0$

Table Bifurcations of the notes

Normal formWhat happens at $r = 0$
$\dot x = r + x^2$ (saddle-node)two equilibria $\pm\sqrt{-r}$ for $r < 0$ merge and vanish
$\dot x = rx - x^2$ (transcritical)$0$ and $r$ exchange stability
$\dot x = rx - x^3$ (pitchfork)$0$ stable for $r < 0$; for $r > 0$, $0$ unstable and $\pm\sqrt r$ stable

Derivation Separable equations

In short: divide by $h(x)$, recognise the chain rule on the left and integrate both sides: $\int dx/h(x) = \int g(t)\,dt$.

Form: $\dot x = g(t)\,h(x)$ (the right-hand side factors into a function of $t$ times a function of $x$).

$\dfrac{1}{h(x)}\dfrac{dx}{dt} = g(t)$
divide where $h(x) \neq 0$; the zeros of $h$ are kept apart
$H(x) = \int\dfrac{dx}{h(x)} \Rightarrow \dfrac{d}{dt}H(x(t)) = H'(x)\,\dot x = \dfrac{\dot x}{h(x)}$
chain rule: this is what makes "separating $dx$ and $dt$" legitimate
$\dfrac{d}{dt}H(x(t)) = g(t) \Rightarrow H(x(t)) = \int g(t)\,dt + C$
integrate both sides in $t$
$H(x) - H(x_0) = \int_{t_0}^{t}g(s)\,ds$
with the initial condition, by Barrow's rule $[F]_{a}^{b} = F(b) - F(a)$

Derivation The logistic equation solved

In short: separate, split $1/(N(K - N))$ by partial fractions, integrate to logarithms and solve for $N$.
$\dfrac{K\,dN}{N(K - N)} = r\,dt$
separation of $\dot N = rN(K - N)/K$
$\dfrac{K}{N(K - N)} = \dfrac1N + \dfrac{1}{K - N}$
partial fractions: $K = (K - N) + N$
$\ln\lvert N\rvert - \ln\lvert K - N\rvert = rt + c$
$\int\frac{dN}{K - N} = -\ln\lvert K - N\rvert$ (inner derivative $-1$)
$\dfrac{N}{K - N} = Ae^{rt},\ A = \dfrac{N_0}{K - N_0}$
exponentiate; $\pm e^c$ and the initial value fix $A$
$N(t) = \dfrac{KN_0}{N_0 + (K - N_0)e^{-rt}}$
solve for $N$, multiply top and bottom by $e^{-rt}$

Limits: $N \to K$ as $t \to \infty$ for any $N_0 > 0$; $N \equiv 0$ and $N \equiv K$ are the equilibria lost when dividing.

Derivation The integrating factor

In short: multiply $\dot x + p(t)x = q(t)$ by $\mu = e^{\int p\,dt}$, so the left side becomes $(\mu x)'$; integrate.
$\mu\dot x + \mu p\,x = \mu q$
multiply by an unknown positive function $\mu(t)$
want $\mu\dot x + \mu p\,x = (\mu x)' = \mu\dot x + \dot\mu x$
product rule backwards: the left side should be one derivative
$\dot\mu = p\mu \Rightarrow \mu(t) = e^{\int p(t)\,dt}$
compare the two left sides; a separable equation for $\mu$ (no constant needed)
$(\mu x)' = \mu q \Rightarrow \mu x = \int\mu q\,dt + C$
integrate both sides
$x(t) = e^{-P(t)}\Big[x_0 + \int_{t_0}^{t}e^{P(s)}q(s)\,ds\Big]$
with $P(t) = \int_{t_0}^{t}p(s)\,ds$, so $P(t_0) = 0$ and Barrow gives $x(t_0) = x_0$

Algorithm Linear first-order IVP in three steps

  1. Put it in standard form $\dot x + p(t)x = q(t)$: divide by the coefficient of $\dot x$ (note where it vanishes: the solution lives on an interval without those points).
  2. $\mu = e^{\int p\,dt}$; check $\dot\mu = p\mu$.
  3. $x = \frac1\mu\big(\int\mu q\,dt + C\big)$; impose $x(t_0) = x_0$ last.

Derivation Exact equations

In short: if $M\,dx + N\,dy = 0$ has $M_y = N_x$, there is $F(x,y)$ with $F_x = M$, $F_y = N$, and the solutions are the level curves $F(x,y) = C$.
$\dfrac{d}{dx}F(x, y(x)) = F_x + F_y\,y'$
chain rule along a solution curve $y(x)$
$= M + N\,y' = 0$
this is the equation itself, so $F$ is constant on solutions
$F_{xy} = F_{yx} \Rightarrow M_y = N_x$
equality of mixed partials (Schwarz): the exactness test
$F = \int M\,dx + g(y)$
integrate $F_x = M$ in $x$; the "constant" may depend on $y$
$F_y = N \Rightarrow g'(y) = N - \partial_y\!\int M\,dx$
the right side must depend on $y$ only (a check of the work)

Derivation Integrating factor for a non-exact equation

In short: look for $\mu(x)$ with $(\mu M)_y = (\mu N)_x$; it exists when $(M_y - N_x)/N$ depends on $x$ only.
$(\mu M)_y = (\mu N)_x$
the new equation $\mu M\,dx + \mu N\,dy = 0$ must be exact
$\mu M_y = \mu' N + \mu N_x$
$\mu = \mu(x)$: no $y$-derivative of $\mu$; product rule in $x$
$\dfrac{\mu'}{\mu} = \dfrac{M_y - N_x}{N}$
rearrange; must be a function of $x$ alone
$\mu(x) = \exp\!\int\dfrac{M_y - N_x}{N}\,dx$
integrate $(\ln\mu)' $
$\mu(y) = \exp\!\int\dfrac{N_x - M_y}{M}\,dy$
the same argument with $\mu = \mu(y)$

Derivation Substitutions: homogeneous and Bernoulli

In short: $y = vx$ turns $y' = F(y/x)$ into a separable equation; $v = y^{1-n}$ turns Bernoulli into a linear one.
$y = vx \Rightarrow y' = v + xv'$
product rule
$v + xv' = F(v) \Rightarrow \dfrac{dv}{F(v) - v} = \dfrac{dx}{x}$
separable in $v$ and $x$
$y' + p(x)y = q(x)y^n,\ n \ne 0, 1$
Bernoulli form
$v = y^{1-n} \Rightarrow v' = (1 - n)y^{-n}y'$
chain rule
$v' + (1 - n)p\,v = (1 - n)q$
multiply Bernoulli by $(1 - n)y^{-n}$: linear in $v$

Algorithm Choosing the method

  1. Autonomous? Phase line first (equilibria, stability), then solve if asked.
  2. Can the right side be written $g(t)h(x)$? Separable.
  3. Linear in $x$: $\dot x + p(t)x = q(t)$? Integrating factor.
  4. Written as $M\,dx + N\,dy = 0$: test $M_y = N_x$; if not, try $\mu(x)$ or $\mu(y)$.
  5. $y' = F(y/x)$: $y = vx$. A power $y^n$ on the right of a linear equation: Bernoulli.
  6. Always add back the constant solutions removed by a division.

Trap Lost and singular solutions

  • Dividing by $h(x)$ removes the equilibria $h(x^*) = 0$: check whether they are solutions and whether the general formula recovers them for some constant.
  • Writing $\ln\lvert x\rvert = \dots$ then $x = Ce^{\dots}$: $C = \pm e^{c}$ can take any sign, and $C = 0$ recovers $x \equiv 0$.
  • Domain: a solution through $(t_0, x_0)$ lives on the largest interval containing $t_0$ where the formula is defined (look for zero denominators and logs of zero).

Unit 3 — Second-order linear ODEs and vibrations

Theorem Existence, uniqueness and the general solution

$\ddot x + p(t)\dot x + q(t)x = f(t)$ with $p, q, f$ continuous on an interval $I$, $t_0 \in I$: for any $x(t_0) = x_0$, $\dot x(t_0) = v_0$ there is exactly one solution, and it exists on all of $I$ (linear equations do not blow up inside $I$).

  • Two conditions are needed because the equation fixes $\ddot x$ once $x$ and $\dot x$ are known.
  • If $x_1, x_2$ solve the homogeneous equation and $W(x_1, x_2) \neq 0$, every solution is $x = C_1x_1 + C_2x_2$; the constants come from
$$\begin{pmatrix} x_1(t_0) & x_2(t_0) \\ \dot x_1(t_0) & \dot x_2(t_0)\end{pmatrix}\begin{pmatrix} C_1 \\ C_2\end{pmatrix} = \begin{pmatrix} x_0 \\ v_0\end{pmatrix},$$

solvable for every datum exactly when the determinant $W(t_0) \neq 0$.

Derivation The Wronskian and Abel's identity

In short: differentiate $W = x_1\dot x_2 - x_2\dot x_1$, use the equation to remove $\ddot x_1$, $\ddot x_2$, and get $\dot W = -pW$.
$\dot W = (\dot x_1\dot x_2 + x_1\ddot x_2) - (\dot x_2\dot x_1 + x_2\ddot x_1)$
product rule on each term
$= x_1\ddot x_2 - x_2\ddot x_1$
the $\dot x_1\dot x_2$ terms cancel
$= x_1(-p\dot x_2 - qx_2) - x_2(-p\dot x_1 - qx_1)$
each $x_i$ solves $\ddot x = -p\dot x - qx$
$= -p\,(x_1\dot x_2 - x_2\dot x_1) = -p\,W$
the $q$ terms cancel
$W(t) = W(t_0)\exp\!\Big(-\int_{t_0}^{t}p(s)\,ds\Big)$
linear first-order equation for $W$ (separable)

Dichotomy: the exponential never vanishes, so for two solutions $W \equiv 0$ on $I$ or $W \neq 0$ everywhere on $I$.

Trap The Wronskian of functions that are not solutions

$W \equiv 0$ implies dependence only for solutions of one linear equation. Counterexample (Peano): $x_1 = t^2$, $x_2 = t\lvert t\rvert$ have $W \equiv 0$ but are independent on $(-1, 1)$; they cannot both solve a second-order linear equation with continuous coefficients there. Conversely, $W(t_0) \neq 0$ at one point always proves independence.

Derivation Constant coefficients: the characteristic equation

In short: try $x = e^{kt}$; every derivative brings out a factor $k$, so $e^{kt}$ solves $\ddot x + p\dot x + qx = 0$ exactly when $k^2 + pk + q = 0$.
$(e^{kt})' = ke^{kt},\ (e^{kt})'' = k^2e^{kt}$
chain rule on $e^{kt}$
$(k^2 + pk + q)\,e^{kt} = 0$
substitute; $e^{kt}$ is a common factor
$k^2 + pk + q = 0,\ k = \dfrac{-p \pm \sqrt{p^2 - 4q}}{2}$
$e^{kt} \neq 0$ never, so the bracket must vanish

Table The three cases

RootsConditionFundamental pair
real $k_1 \neq k_2$$p^2 > 4q$$e^{k_1t},\ e^{k_2t}$
double $k = -p/2$$p^2 = 4q$$e^{kt},\ t\,e^{kt}$
$\rho \pm \omega i$$p^2 < 4q$$e^{\rho t}\cos\omega t,\ e^{\rho t}\sin\omega t$

With $\rho = -p/2$ and $\omega = \sqrt{4q - p^2}/2$. Higher order: a root of multiplicity $m$ gives $e^{kt}, te^{kt}, \dots, t^{m-1}e^{kt}$.

Derivation Why $te^{kt}$ for a double root

In short: substituting $te^{kt}$ leaves $[(k^2 + pk + q)t + (2k + p)]e^{kt}$, and both brackets vanish at a double root.
$(te^{kt})' = (1 + kt)e^{kt}$
product rule
$(te^{kt})'' = (2k + k^2t)e^{kt}$
product rule again
$L[te^{kt}] = \big[(k^2 + pk + q)t + (2k + p)\big]e^{kt}$
collect powers of $t$
$k^2 + pk + q = 0$ and $2k + p = 0$
a double root of $P(k)$ is also a root of $P'(k) = 2k + p$

Derivation Real solutions from complex roots

In short: Euler's formula splits $e^{(\rho + \omega i)t}$ into $e^{\rho t}(\cos\omega t + i\sin\omega t)$; real and imaginary parts are real solutions.
$e^{(\rho \pm \omega i)t} = e^{\rho t}(\cos\omega t \pm i\sin\omega t)$
Euler: $e^{i\theta} = \cos\theta + i\sin\theta$; $\cos$ is even, $\sin$ odd
$\tfrac12(x_+ + x_-) = e^{\rho t}\cos\omega t$
superposition with $C_1 = C_2 = \tfrac12$
$\tfrac{1}{2i}(x_+ - x_-) = e^{\rho t}\sin\omega t$
superposition with $C_{1,2} = \pm\tfrac1{2i}$
$x = e^{\rho t}(A\cos\omega t + B\sin\omega t)$
general real solution, $A, B$ real

Derivation Amplitude and phase

In short: $A\cos\omega t + B\sin\omega t = M\cos(\omega t - \phi)$ with $M = \sqrt{A^2 + B^2}$, $\cos\phi = A/M$, $\sin\phi = B/M$.
$M\cos(\omega t - \phi) = M\cos\phi\cos\omega t + M\sin\phi\sin\omega t$
$\cos(a - b) = \cos a\cos b + \sin a\sin b$
$A = M\cos\phi,\ B = M\sin\phi$
match the coefficients of $\cos\omega t$ and $\sin\omega t$
$M^2 = A^2 + B^2$
square and add, $\cos^2 + \sin^2 = 1$

Trap: $\tan\phi = B/A$ alone loses the quadrant; read it from the signs of $\cos\phi = A/M$ and $\sin\phi = B/M$.

Table Mass-spring-damper $\ddot x + \lambda\dot x + \omega^2x = 0$

$\lambda = \mu/m$ (damping per unit mass), $\omega^2 = k/m$ (natural frequency squared). Roots $-\lambda/2 \pm \sqrt{\lambda^2/4 - \omega^2}$.

RegimeConditionMotion
undamped$\lambda = 0$$M\cos(\omega t - \phi)$, period $2\pi/\omega$
underdamped$\lambda < 2\omega$$Me^{-\lambda t/2}\cos(\sigma t - \phi)$, $\sigma = \sqrt{\omega^2 - \lambda^2/4}$
critical$\lambda = 2\omega$$(A + Bt)e^{-\lambda t/2}$: fastest return without oscillation
overdamped$\lambda > 2\omega$$Ae^{k_1t} + Be^{k_2t}$, $k_{1,2} < 0$

$\sigma < \omega$: damping slows the oscillation; the envelope $\pm Me^{-\lambda t/2}$ bounds it. Over- and critical damping cross zero at most once.

Derivation Structure of the inhomogeneous solution

In short: the difference of two solutions of $L[x] = f$ solves $L[x] = 0$, so $x = x_H + x_N$.
$L[x - x_N] = L[x] - L[x_N] = f - f = 0$
linearity of $L$
$x - x_N = C_1x_1 + C_2x_2$
general solution of the homogeneous equation
$x = C_1x_1 + C_2x_2 + x_N$
impose the initial conditions on this sum, never on $x_H$ alone

Table The educated guess (undetermined coefficients)

Forcing $f(t)$Try $x_N$
polynomial of degree $n$general polynomial of degree $n$
$ae^{\alpha t}$$Ce^{\alpha t}$
$a\cos\alpha t$ or $a\sin\alpha t$$C\cos\alpha t + D\sin\alpha t$
$P_n(t)e^{\alpha t}\cos\beta t$$e^{\alpha t}\big(Q_n(t)\cos\beta t + R_n(t)\sin\beta t\big)$
sum $f_1 + f_2$sum of the two guesses (superposition)

Resonance rule: if the exponent of the forcing ($\alpha$, $i\alpha$ or $\alpha + i\beta$) is a root of the characteristic equation of multiplicity $s$, multiply the guess by $t^s$. Example of the notes: $\ddot x + \dot x - 6x = 5e^{-3t}$ with $k = -3$ a simple root gives $x_N = Cte^{-3t}$, $C = -1$.

Derivation Undamped resonance

In short: for $\ddot x + \omega^2x = a\cos\omega t$ the guess $t(C\cos\omega t + D\sin\omega t)$ gives $x_N = \dfrac{a}{2\omega}t\sin\omega t$, an amplitude growing linearly.
$x_N = t\,g(t),\ g = C\cos\omega t + D\sin\omega t$
$i\omega$ is a simple root of $k^2 + \omega^2$: multiply by $t$
$\ddot x_N = 2\dot g + t\ddot g = 2\dot g - \omega^2tg$
product rule twice; $\ddot g = -\omega^2g$
$\ddot x_N + \omega^2x_N = 2\dot g = 2\omega(D\cos\omega t - C\sin\omega t)$
the $t$ terms cancel
$C = 0,\ D = \dfrac{a}{2\omega}$
match with $a\cos\omega t$

Off resonance ($\alpha \neq \omega$): $x_N = \dfrac{a\cos\alpha t}{\omega^2 - \alpha^2}$. Starting from rest, $x = \dfrac{2a}{\omega^2 - \alpha^2}\sin\dfrac{(\omega - \alpha)t}{2}\sin\dfrac{(\omega + \alpha)t}{2}$: beats, a fast oscillation inside a slow envelope.

Derivation Damped forced response

In short: $\ddot x + \lambda\dot x + \omega^2x = a\cos\alpha t$ has the steady state $R\cos(\alpha t - \delta)$ with $R = a/\sqrt{(\omega^2 - \alpha^2)^2 + \lambda^2\alpha^2}$.
$x_N = C\sin\alpha t + D\cos\alpha t$
$i\alpha$ is not a root when $\lambda > 0$
$C(\omega^2 - \alpha^2) = D\lambda\alpha,\ D(\omega^2 - \alpha^2) + C\lambda\alpha = a$
substitute and match $\sin\alpha t$, $\cos\alpha t$
$C = \dfrac{a\lambda\alpha}{\Delta},\ D = \dfrac{a(\omega^2 - \alpha^2)}{\Delta}$
Cramer, $\Delta = (\omega^2 - \alpha^2)^2 + \lambda^2\alpha^2$
$R = \sqrt{C^2 + D^2} = a/\sqrt{\Delta}$
amplitude; $\tan\delta = \lambda\alpha/(\omega^2 - \alpha^2)$
$\alpha_r = \sqrt{\omega^2 - \lambda^2/2}$
minimise $\Delta$ in $s = \alpha^2$: $-2(\omega^2 - s) + \lambda^2 = 0$; exists if $\lambda^2 < 2\omega^2$

The homogeneous part decays like $e^{-\lambda t/2}$ (transient); only $R\cos(\alpha t - \delta)$ remains (steady state).

Derivation Reduction of order

In short: if $u$ is one solution, $x = uy$ makes the equation first order in $z = \dot y$, giving $x_2 = u\int\dfrac{e^{-\int p\,dt}}{u^2}\,dt$.
$\dot x = \dot uy + u\dot y,\ \ddot x = \ddot uy + 2\dot u\dot y + u\ddot y$
product rule
$[\ddot u + p\dot u + qu]\,y + [2\dot u + pu]\,\dot y + u\ddot y = 0$
substitute and group by $y$, $\dot y$, $\ddot y$
$u\dot z + (2\dot u + pu)z = 0,\ z = \dot y$
the first bracket is zero because $u$ is a solution; no $y$ left
$\dfrac{\dot z}{z} = -\dfrac{2\dot u}{u} - p \Rightarrow z = \dfrac{A\,e^{-\int p\,dt}}{u^2}$
separable; $\int 2\dot u/u = 2\ln\lvert u\rvert$
$x_2 = u\int z\,dt$
$y = \int z$; take $A = 1$ and no constant (it only adds $u$ again)

For the inhomogeneous equation the same substitution gives the linear equation $u\dot z + (2\dot u + pu)z = f$.

Derivation Variation of constants

In short: look for $x_N = u_1x_1 + u_2x_2$ with $\dot u_1x_1 + \dot u_2x_2 = 0$; then $\dot u_1 = -x_2f/W$, $\dot u_2 = x_1f/W$.
$\dot x_N = u_1\dot x_1 + u_2\dot x_2 + (\dot u_1x_1 + \dot u_2x_2)$
product rule
$\dot u_1x_1 + \dot u_2x_2 = 0$
free choice (two unknowns, one equation so far): keeps $\ddot x_N$ free of $\ddot u_i$
$L[x_N] = u_1L[x_1] + u_2L[x_2] + \dot u_1\dot x_1 + \dot u_2\dot x_2 = f$
differentiate again and substitute; $L[x_i] = 0$
$\begin{pmatrix} x_1 & x_2 \\ \dot x_1 & \dot x_2\end{pmatrix}\begin{pmatrix}\dot u_1 \\ \dot u_2\end{pmatrix} = \begin{pmatrix} 0 \\ f\end{pmatrix}$
the two conditions together; the determinant is $W$
$x_N = -x_1\int\dfrac{x_2f}{W}\,dt + x_2\int\dfrac{x_1f}{W}\,dt$
Cramer's rule, then integrate

$f$ is the forcing of the normalised equation (coefficient of $\ddot x$ equal to $1$).

Derivation Euler-Cauchy equations

In short: $t^2\ddot x + at\dot x + bx = 0$ becomes constant-coefficient with $t = e^s$ ($t > 0$), or directly with $x = t^m$: $m(m - 1) + am + b = 0$.
$\dfrac{dx}{dt} = \dfrac{dx}{ds}\dfrac{ds}{dt} = \dfrac1t\,x_s$
chain rule, $s = \ln t$
$\dfrac{d^2x}{dt^2} = -\dfrac{1}{t^2}x_s + \dfrac{1}{t^2}x_{ss}$
product rule on $\frac1t x_s$, chain rule again
$x_{ss} + (a - 1)x_s + bx = 0$
multiply by $t^2$ and substitute
$x = t^m$: $m(m - 1) + am + b = 0$
same as $e^{ms}$ in the $s$-equation

Double root $m$: $t^m$, $t^m\ln t$. Complex $m = \alpha \pm i\beta$: $t^\alpha\cos(\beta\ln t)$, $t^\alpha\sin(\beta\ln t)$.

Algorithm Second-order IVP from start to finish

  1. Normalise: divide by the coefficient of $\ddot x$; note where $p$, $q$, $f$ are continuous (the interval of the solution).
  2. Homogeneous part: constant coefficients, characteristic equation; Euler-Cauchy, $t^m$; one solution known, reduction of order.
  3. Particular part: guess (with the resonance rule) for exponential, polynomial or trigonometric forcing; otherwise variation of constants.
  4. $x = C_1x_1 + C_2x_2 + x_N$; differentiate the whole sum.
  5. Impose $x(t_0)$ and $\dot x(t_0)$; solve the $2\times 2$ system.
  6. Check: substitute back, check the initial values, units and the limit $t \to \infty$.

Trap Second-order exam traps

  • Initial conditions imposed on $x_H$ before adding $x_N$.
  • Forgetting the factor $t^s$ when the forcing resonates with a root.
  • Using variation of constants with $f$ of the non-normalised equation.
  • $e^{kt}$ and $te^{kt}$ written for complex roots, or $\cos$ and $\sin$ without the factor $e^{\rho t}$.

Unit 4 — Linear systems and the phase plane

Pattern From an $n$-th order equation to a system

Set $x_1 = x$, $x_2 = \dot x$, ..., $x_n = x^{(n-1)}$. Then $\dot x_1 = x_2$, ..., $\dot x_n = $ the equation solved for $x^{(n)}$. The damped oscillator becomes

$$\frac{d}{dt}\begin{pmatrix} x_1 \\ x_2\end{pmatrix} = \begin{pmatrix} 0 & 1 \\ -\omega^2 & -\lambda\end{pmatrix}\begin{pmatrix} x_1 \\ x_2\end{pmatrix}.$$

Its trace is $-\lambda$ and its determinant $\omega^2$: the characteristic equations of the scalar equation and of the system coincide.

Theorem Linear systems: structure and Liouville

$\dot{\mathbf x} = A(t)\mathbf x + \mathbf f(t)$ with continuous entries has a unique solution on the whole interval for each $\mathbf x(t_0)$.

  • $n$ independent solutions form the fundamental matrix $\Phi = [\mathbf x_1 \mid \cdots \mid \mathbf x_n]$; general solution $\mathbf x = \Phi\mathbf c + \mathbf x_p$.
  • Liouville: $W = \det\Phi$ satisfies $\dot W = (\operatorname{tr}A)W$, so $W(t) = W(t_0)\exp\int_{t_0}^{t}\operatorname{tr}A(s)\,ds$: zero everywhere or nowhere.
  • Variation of constants: $\mathbf x_p = \Phi\int\Phi^{-1}\mathbf f\,dt$, from $\mathbf x = \Phi\mathbf c(t)$: $\dot\Phi\mathbf c + \Phi\dot{\mathbf c} = A\Phi\mathbf c + \mathbf f$ and $\dot\Phi = A\Phi$ leave $\Phi\dot{\mathbf c} = \mathbf f$.

Derivation Elimination

In short: from $\dot x = ax + by$, $\dot y = cx + dy$ with $b \neq 0$, eliminating $y$ gives $\ddot x - (\operatorname{tr}A)\dot x + (\det A)x = 0$.
$y = (\dot x - ax)/b$
first equation solved for $y$
$\ddot x = a\dot x + b\dot y = a\dot x + b(cx + dy)$
differentiate the first equation, insert the second
$\ddot x = a\dot x + bcx + d(\dot x - ax)$
replace $by = \dot x - ax$
$\ddot x - (a + d)\dot x + (ad - bc)x = 0$
collect: a Unit 3 equation
$y = (\dot x - ax)/b$
recover $y$ by differentiating $x$; never integrate a second time

Derivation The eigenvalue method

In short: try $\mathbf x = e^{\lambda t}\mathbf v$; it solves $\dot{\mathbf x} = A\mathbf x$ exactly when $A\mathbf v = \lambda\mathbf v$, so $\lambda^2 - \tau\lambda + \Delta = 0$.
$\dot{\mathbf x} = \lambda e^{\lambda t}\mathbf v,\ A\mathbf x = e^{\lambda t}A\mathbf v$
$\mathbf v$ is constant
$(A - \lambda I)\mathbf v = \mathbf 0,\ \mathbf v \neq \mathbf 0$
equate and divide by $e^{\lambda t} \neq 0$
$\det(A - \lambda I) = 0$
a homogeneous system has non-zero solutions only if singular
$\lambda^2 - \tau\lambda + \Delta = 0,\ \tau = \operatorname{tr}A,\ \Delta = \det A$
expand the $2\times 2$ determinant
$\lambda = \dfrac{\tau}{2} \pm \sqrt{\dfrac{\tau^2}{4} - \Delta}$
quadratic formula; $\lambda_1 + \lambda_2 = \tau$, $\lambda_1\lambda_2 = \Delta$

Derivation Distinct real eigenvalues

In short: $\mathbf x = C_1e^{\lambda_1t}\mathbf v_1 + C_2e^{\lambda_2t}\mathbf v_2$; in $\mathbf z = P^{-1}\mathbf x$ with $P = [\mathbf v_1 \mid \mathbf v_2]$ the system decouples and orbits are $z_2 = Kz_1^{\lambda_2/\lambda_1}$.
$AP = [\lambda_1\mathbf v_1 \mid \lambda_2\mathbf v_2] = PD$
column by column, $A\mathbf v_i = \lambda_i\mathbf v_i$
$\dot{\mathbf z} = P^{-1}AP\mathbf z = D\mathbf z$
$\mathbf x = P\mathbf z$, $P$ constant
$z_1 = C_1e^{\lambda_1t},\ z_2 = C_2e^{\lambda_2t}$
two uncoupled decay or growth laws
$z_2 = C_2\big(e^{\lambda_1t}\big)^{\lambda_2/\lambda_1} = Kz_1^{\lambda_2/\lambda_1}$
eliminate $t$: the orbits
$\mathbf x = P\mathbf z = C_1e^{\lambda_1t}\mathbf v_1 + C_2e^{\lambda_2t}\mathbf v_2$
back to the original variables

Node ($\lambda_1, \lambda_2$ same sign): orbits leave or enter tangent to the eigenvector of smaller $\lvert\lambda\rvert$ (the slow direction) and become parallel to the fast one far away. Saddle (opposite signs): the line of $\mathbf v$ with $\lambda < 0$ is the stable manifold, the one with $\lambda > 0$ the unstable manifold. Zero eigenvalue: $\mathbf x = C_1\mathbf v_1 + C_2e^{\lambda_2t}\mathbf v_2$, a line of equilibria along $\mathbf v_1$ and straight orbits parallel to $\mathbf v_2$.

Derivation Complex eigenvalues: canonical form

In short: with $\lambda = \rho + i\omega$, $\mathbf v = \mathbf u + i\mathbf w$ and $Q = [\mathbf u \mid \mathbf w]$, $Q^{-1}AQ = J = \begin{pmatrix}\rho & \omega \\ -\omega & \rho\end{pmatrix}$.
$A(\mathbf u + i\mathbf w) = (\rho + i\omega)(\mathbf u + i\mathbf w)$
eigenvector for $\rho + i\omega$ (not $\rho - i\omega$)
$A\mathbf u = \rho\mathbf u - \omega\mathbf w,\ A\mathbf w = \omega\mathbf u + \rho\mathbf w$
expand, $i^2 = -1$; real and imaginary parts separately
$AQ = [\mathbf u \mid \mathbf w]\begin{pmatrix}\rho & \omega \\ -\omega & \rho\end{pmatrix} = QJ$
the two lines above, written column by column
$\dot z_1 = \rho z_1 + \omega z_2,\ \dot z_2 = -\omega z_1 + \rho z_2$
$\mathbf z = Q^{-1}\mathbf x$
$r = r_0e^{\rho t},\ \theta = \theta_0 - \omega t$
polar coordinates: $\dot r = \rho r$, $\dot\theta = -\omega$ (clockwise in the $\mathbf z$-plane)

Real general solution:

$$\mathbf x = e^{\rho t}\big[(C_1\cos\omega t + C_2\sin\omega t)\mathbf u + (C_2\cos\omega t - C_1\sin\omega t)\mathbf w\big],$$

that is $C_1\operatorname{Re}(\mathbf ve^{\lambda t}) + C_2\operatorname{Im}(\mathbf ve^{\lambda t})$.

Interpretation Spirals and centres

  • $\rho < 0$: stable spiral (sink); $\rho > 0$: unstable spiral (source); $\rho = 0$: centre, closed ellipses of period $2\pi/\omega$.
  • Every turn multiplies the distance by $e^{2\pi\rho/\omega}$.
  • Sense of rotation in the $\mathbf x$-plane: compute $\dot{\mathbf x} = A\mathbf x$ at $\mathbf x = (1, 0)^T$: it is $(a, c)^T$. If $c > 0$ the orbit turns counterclockwise, if $c < 0$ clockwise (the sense is the sign of the entry $c$ of $A$).

Derivation Repeated eigenvalue: stars and improper nodes

In short: if $A \neq \lambda I$, take $\mathbf v_1$ with $(A - \lambda I)\mathbf v_1 = 0$ and $\mathbf v_2$ with $(A - \lambda I)\mathbf v_2 = \mathbf v_1$; then $\mathbf x = e^{\lambda t}[c_1\mathbf v_1 + c_2(t\mathbf v_1 + \mathbf v_2)]$.
$\lambda = \tau/2$, $\tau^2 = 4\Delta$
double root of $\lambda^2 - \tau\lambda + \Delta$
$A = \lambda I$: every vector is an eigenvector
geometric multiplicity 2: star, $\mathbf x = e^{\lambda t}\mathbf x_0$, straight rays
$A\mathbf v_2 = \lambda\mathbf v_2 + \mathbf v_1 \Rightarrow Q^{-1}AQ = \begin{pmatrix}\lambda & 1 \\ 0 & \lambda\end{pmatrix}$
$Q = [\mathbf v_1 \mid \mathbf v_2]$, column by column
$\dot z_2 = \lambda z_2 \Rightarrow z_2 = c_2e^{\lambda t}$
the second row is uncoupled
$\dot z_1 = \lambda z_1 + c_2e^{\lambda t} \Rightarrow (e^{-\lambda t}z_1)' = c_2$
integrating factor $e^{-\lambda t}$
$z_1 = (c_1 + c_2t)e^{\lambda t}$
integrate, multiply back by $e^{\lambda t}$

Improper node: all orbits are tangent to $\mathbf v_1$ at the origin (for $\lambda < 0$, as $t \to +\infty$). $\mathbf v_2$ is not unique: $\mathbf v_2 + \alpha\mathbf v_1$ works too.

Table Classification by trace and determinant

ConditionType of the origin
$\Delta < 0$saddle (always unstable)
$\Delta > 0$, $\tau^2 > 4\Delta$node: stable if $\tau < 0$, unstable if $\tau > 0$
$\Delta > 0$, $\tau^2 = 4\Delta$star ($A = \lambda I$) or improper node
$\Delta > 0$, $\tau^2 < 4\Delta$, $\tau \neq 0$spiral: stable if $\tau < 0$
$\Delta > 0$, $\tau = 0$centre
$\Delta = 0$a line of equilibria (zero eigenvalue)

Prerequisite Matrix exponential (complement)

The fundamental matrix with $\Phi(0) = I$ is $e^{At}$, and $\mathbf x(t) = e^{At}\mathbf x_0$. From the canonical forms:

  • $A = PDP^{-1}$: $e^{At} = P\operatorname{diag}(e^{\lambda_1t}, e^{\lambda_2t})P^{-1}$.
  • $J = \begin{pmatrix}\lambda & 1 \\ 0 & \lambda\end{pmatrix}$: $e^{Jt} = e^{\lambda t}\begin{pmatrix} 1 & t \\ 0 & 1\end{pmatrix}$.
  • $J = \begin{pmatrix}\rho & \omega \\ -\omega & \rho\end{pmatrix}$: $e^{Jt} = e^{\rho t}\begin{pmatrix}\cos\omega t & \sin\omega t \\ -\sin\omega t & \cos\omega t\end{pmatrix}$.
  • Then $e^{At} = Qe^{Jt}Q^{-1}$. This packaging is an extension; the notes reach the same solutions through $\mathbf z = Q^{-1}\mathbf x$.

Algorithm Planar system from start to finish

  1. $\tau = \operatorname{tr}A$, $\Delta = \det A$; classify the origin with the table before computing anything.
  2. Eigenvalues from $\lambda^2 - \tau\lambda + \Delta = 0$.
  3. Eigenvectors: for the row $(a - \lambda,\ b)$ with $b \neq 0$ take $\mathbf v = (b,\ \lambda - a)^T$.
  4. Build $P$ or $Q$ and check $Q^{-1}AQ$ equals the expected $D$ or $J$.
  5. Write the general solution; impose $\mathbf x(0) = \mathbf x_0$ by solving $Q\mathbf c = \mathbf x_0$.
  6. Phase portrait: eigenvector lines, slow and fast directions, sense of rotation from $\dot{\mathbf x}$ at one point.
  7. Constant forcing $\dot{\mathbf x} = A\mathbf x + \mathbf b$: equilibrium $\mathbf x^* = -A^{-1}\mathbf b$, then $\mathbf y = \mathbf x - \mathbf x^*$ solves $\dot{\mathbf y} = A\mathbf y$.

Trap Phase-plane traps

  • Taking the eigenvector of $\rho - i\omega$: $J$ comes out with $\omega$ and $-\omega$ swapped and the rotation reversed.
  • Forgetting that eigenvectors can be scaled: different-looking answers may be the same.
  • Calling a repeated eigenvalue with $A \neq \lambda I$ a star: it is an improper node with only one eigenvector line.
  • Reading a node's tangent direction from the fast eigenvector instead of the slow one.

Units 5 and 6 — Nonlinear systems, Fourier series and PDEs (extension)

No class notes for these units are in the repository yet. The cards below collect the standard results used in the solved Extra problems P08 to P22; check them against the official notes when they arrive.

Algorithm Nonlinear planar systems

  1. Equilibria: solve $f(x, y) = 0$, $g(x, y) = 0$ together (factor first; check every branch).
  2. Jacobian $D\mathbf F = \begin{pmatrix} f_x & f_y \\ g_x & g_y\end{pmatrix}$ at each equilibrium.
  3. Classify each Jacobian with the trace-determinant table of Unit 4.
  4. Linearisation theorem: if no eigenvalue has zero real part (hyperbolic point), the nonlinear system has the same type nearby. A linear centre is inconclusive.
  5. Nullclines $f = 0$ (vertical flow) and $g = 0$ (horizontal flow) to join the local pictures.

Derivation Conservative systems and energy

In short: for $\ddot x = -V'(x)$, $E = \tfrac12\dot x^2 + V(x)$ is constant, so orbits are level curves of $E$: minima of $V$ are centres, maxima saddles.
$\dot x = y,\ \dot y = -V'(x)$
first-order system
$\dfrac{dE}{dt} = E_x\dot x + E_y\dot y = V'(x)y + y(-V'(x)) = 0$
chain rule along an orbit
$y = \pm\sqrt{2(E - V(x))}$
each level curve, real where $V(x) \le E$

This settles the centres that the linearisation leaves undecided.

Table Fourier series on $(-L, L)$

ItemFormula
Series$f \sim \dfrac{a_0}{2} + \sum_{n\ge 1}\Big(a_n\cos\dfrac{n\pi x}{L} + b_n\sin\dfrac{n\pi x}{L}\Big)$
Cosine coefficients$a_n = \dfrac1L\int_{-L}^{L}f(x)\cos\dfrac{n\pi x}{L}\,dx$
Sine coefficients$b_n = \dfrac1L\int_{-L}^{L}f(x)\sin\dfrac{n\pi x}{L}\,dx$
$f$ even$b_n = 0$, $a_n = \frac2L\int_0^L f\cos\frac{n\pi x}{L}\,dx$
$f$ odd$a_n = 0$, $b_n = \frac2L\int_0^L f\sin\frac{n\pi x}{L}\,dx$
Half range on $(0, L)$sine series = odd extension, cosine series = even extension
Convergence (Dirichlet)piecewise smooth $f$: the series gives $\tfrac12\big(f(x^+) + f(x^-)\big)$, also at $\pm L$ via the periodic extension

With $L = \pi$ the arguments are $nx$. Evaluating the series at a point where $f$ is known gives sums such as $\sum 1/n^2 = \pi^2/6$ and $\sum (-1)^{n+1}/n^2 = \pi^2/12$.

Derivation Why the coefficient formulas work

In short: multiply the series by $\cos(m\pi x/L)$ and integrate; orthogonality kills every term except $n = m$.
$\cos a\cos b = \tfrac12[\cos(a - b) + \cos(a + b)]$
product-to-sum
$\int_{-L}^{L}\cos\dfrac{n\pi x}{L}\cos\dfrac{m\pi x}{L}\,dx = L\,\delta_{nm}$
each cosine of a non-zero multiple integrates to $0$ over a full period ($n, m \ge 1$)
$\int_{-L}^{L}\sin\dfrac{n\pi x}{L}\cos\dfrac{m\pi x}{L}\,dx = 0$
odd integrand on a symmetric interval
$\int_{-L}^{L}f\cos\dfrac{m\pi x}{L}\,dx = a_mL$
only the $n = m$ term survives

Table Eigenvalue problems $X'' + \lambda X = 0$ on $(0, L)$

Boundary conditionsEigenvaluesEigenfunctions
$X(0) = X(L) = 0$$(n\pi/L)^2,\ n \ge 1$$\sin(n\pi x/L)$
$X'(0) = X'(L) = 0$$(n\pi/L)^2,\ n \ge 0$$\cos(n\pi x/L)$, with the constant for $n = 0$
$X(0) = 0,\ X'(L) = 0$$\big((2n - 1)\pi/(2L)\big)^2$$\sin\big((2n - 1)\pi x/(2L)\big)$
Robin, e.g. $X'(L) + hX(L) = 0$roots of a transcendental equation, e.g. $\tan\sqrt\lambda L = -\sqrt\lambda/h$$\sin\sqrt\lambda x$

Always test $\lambda < 0$, $\lambda = 0$ and $\lambda > 0$ separately: a zero mode appears only with Neumann conditions at both ends.

Algorithm Separation of variables

  1. Homogeneous boundary conditions first: subtract a steady state $u_s(x)$ that carries the non-zero boundary values or the source.
  2. $u = X(x)T(t)$; divide by $XT$; both sides equal a constant $-\lambda$.
  3. Solve the eigenvalue problem for $X$ (table above).
  4. Time (or $y$) factor: heat $T' + k\lambda T = 0 \Rightarrow e^{-k\lambda t}$; wave $T'' + c^2\lambda T = 0 \Rightarrow \cos\omega_nt$, $\sin\omega_nt$ with $\omega_n = c\sqrt{\lambda_n}$; Laplace $Y'' - \lambda Y = 0 \Rightarrow \cosh$, $\sinh$ (and $A_0 + B_0y$ for $\lambda = 0$).
  5. Superpose all modes and fit the remaining data with the Fourier coefficient formulas.
  6. Checks: boundary values, initial values, $t \to \infty$ (heat tends to $u_s$), units of $k$ and $c$.

Table The three classical equations on $(0, L)$ with $u = 0$ at both ends

EquationSolution
Heat $u_t = ku_{xx}$$\sum c_n\sin\dfrac{n\pi x}{L}\,e^{-kn^2\pi^2t/L^2}$, $c_n = \frac2L\int_0^L u(x,0)\sin\frac{n\pi x}{L}\,dx$
Wave $u_{tt} = c^2u_{xx}$$\sum\big(A_n\cos\omega_nt + B_n\sin\omega_nt\big)\sin\dfrac{n\pi x}{L}$, $\omega_n = \frac{n\pi c}{L}$; $A_n$ from $u(x,0)$, $\omega_nB_n$ from $u_t(x,0)$
Laplace in a rectangle$\sum\sin\dfrac{n\pi x}{L}\Big(A_n\cosh\dfrac{n\pi y}{L} + B_n\sinh\dfrac{n\pi y}{L}\Big)$
Laplace in a disc or wedge$r^n$ and $r^{-n}$ times $\cos n\theta$, $\sin n\theta$; keep only $r^n$ if bounded at $r = 0$

The slowest heat mode decays with the time constant $\tau_1 = L^2/(k\pi^2)$. The wave equation also has d'Alembert's form $u = \tfrac12[F(x - ct) + F(x + ct)] + \frac{1}{2c}\int_{x - ct}^{x + ct}g(s)\,ds$.

Pocket prerequisites

Prerequisite Integrals that keep coming back

IntegralResult
$\int\dfrac{f'(x)}{f(x)}\,dx$$\ln\lvert f(x)\rvert + C$
$\int\dfrac{dx}{a^2 + x^2}$$\dfrac1a\arctan\dfrac xa + C$
$\int\dfrac{dx}{x^2 - a^2}$$\dfrac{1}{2a}\ln\left\lvert\dfrac{x - a}{x + a}\right\rvert + C$
$\int te^{at}\,dt$$e^{at}\Big(\dfrac ta - \dfrac{1}{a^2}\Big) + C$
$\int e^{at}\cos bt\,dt$$\dfrac{e^{at}(a\cos bt + b\sin bt)}{a^2 + b^2} + C$
$\int e^{at}\sin bt\,dt$$\dfrac{e^{at}(a\sin bt - b\cos bt)}{a^2 + b^2} + C$
$\int x\cos nx\,dx$$\dfrac{x\sin nx}{n} + \dfrac{\cos nx}{n^2} + C$
$\int x\sin nx\,dx$$-\dfrac{x\cos nx}{n} + \dfrac{\sin nx}{n^2} + C$
$\int x^2\cos nx\,dx$$\dfrac{x^2\sin nx}{n} + \dfrac{2x\cos nx}{n^2} - \dfrac{2\sin nx}{n^3} + C$
$\int\tan x\,dx$$-\ln\lvert\cos x\rvert + C$

Integration by parts: $\int u\,dv = uv - \int v\,du$; with limits $\int_a^b u\,dv = [uv]_a^b - \int_a^b v\,du$.

Prerequisite Values at multiples of $\pi$

  • $\sin n\pi = 0$, $\cos n\pi = (-1)^n$, $1 - (-1)^n = 0$ for even $n$ and $2$ for odd $n$.
  • $\sin\dfrac{n\pi}{2} = 0, 1, 0, -1, \dots$ for $n = 0, 1, 2, 3$; $\cos\dfrac{(2k + 1)\pi}{2} = 0$.

Prerequisite Partial fractions

  • Distinct linear factors: $\dfrac{1}{(x - a)(x - b)} = \dfrac{1}{a - b}\Big(\dfrac{1}{x - a} - \dfrac{1}{x - b}\Big)$.
  • Logistic: $\dfrac{K}{x(K - x)} = \dfrac1x + \dfrac{1}{K - x}$.
  • Repeated factor: $\dfrac{A}{x - a} + \dfrac{B}{(x - a)^2}$. Irreducible quadratic: $\dfrac{Ax + B}{x^2 + px + q}$.
  • Cover-up rule: the coefficient of $\frac{1}{x - a}$ is the rest of the fraction evaluated at $x = a$.

Prerequisite Trigonometric and hyperbolic identities

IdentityUse
$e^{i\theta} = \cos\theta + i\sin\theta$complex roots and eigenvalues
$\cos^2x = \tfrac12(1 + \cos 2x),\ \sin^2x = \tfrac12(1 - \cos 2x)$integrals of squares, energies
$\sin a\cos b = \tfrac12[\sin(a + b) + \sin(a - b)]$Fourier coefficients
$\cos a - \cos b = -2\sin\tfrac{a + b}{2}\sin\tfrac{a - b}{2}$beats
$\cosh x = \tfrac12(e^x + e^{-x}),\ \sinh x = \tfrac12(e^x - e^{-x})$real roots $\pm k$, Laplace modes
$\cosh^2x - \sinh^2x = 1,\ (\cosh x)' = \sinh x,\ (\sinh x)' = \cosh x$checks of hyperbolic solutions

Prerequisite $2\times 2$ linear algebra

  • $A = \begin{pmatrix} a & b \\ c & d\end{pmatrix}$: $\tau = a + d$, $\Delta = ad - bc$, $A^{-1} = \dfrac1\Delta\begin{pmatrix} d & -b \\ -c & a\end{pmatrix}$.
  • Eigenvalues: $\lambda = \tfrac\tau2 \pm \sqrt{\tfrac{\tau^2}{4} - \Delta}$; $\lambda_1 + \lambda_2 = \tau$, $\lambda_1\lambda_2 = \Delta$.
  • Eigenvector shortcut: $\mathbf v = (b,\ \lambda - a)^T$ if $b \neq 0$, or $(\lambda - d,\ c)^T$ if $c \neq 0$; check with the first row: $(a - \lambda)b + b(\lambda - a) = 0$.
  • Always verify $Q^{-1}AQ$ by multiplying out: it catches swapped columns and sign slips.

Prerequisite Complex numbers and logarithms

  • $z = a + ib = re^{i\theta}$, $r = \sqrt{a^2 + b^2}$, $\theta$ from $\cos\theta = a/r$, $\sin\theta = b/r$ (mind the quadrant).
  • $\operatorname{Re}(e^{(\rho + i\omega)t}) = e^{\rho t}\cos\omega t$, $\operatorname{Im}(e^{(\rho + i\omega)t}) = e^{\rho t}\sin\omega t$.
  • $e^{\ln x} = x$ for $x > 0$; $\ln\lvert x\rvert = g + c \Rightarrow x = Ce^{g}$ with $C = \pm e^{c}$, and $C = 0$ when $x \equiv 0$ is a solution.