MMP cheatsheet - PH EPFL, 2021 Common eigenbasis: A and B self Wronskian W (y1 , ..., yn )(x) • s1 = s2∗ : 2 lin. indep.

• s1 = s2∗ : 2 lin. indep. sol.

Author: Xavier KERVYN, Page 1 /2 adjoint have common orthonormal  y (Re/Im) 3 Fourier Series
eigenbasis ⇔ [A, B] = 0. ⇒ ∃U unita-  10 ... yn  Hilbert space L2ω ([a, b]) → orthogo-
1 Function & Hilbert spaces
ry such that U† AU = DA and U† BU =  y1 ... yn0  • s1 = s2 ∈ R: always 1st sol.
y (x), nal basis φn (x) = einx .
Inner product space DB both diagonal. W =  .. .. ..  1
Vector space V over field C  . . .  ∞ ∞
Dirac Bra-Ket  (n−1) (n−1)  X
bj (x − x0 )j+s
w. (sequilinear) scalar product:  y . . . yn
X
1 y2 = y1 ln |x − x0 | + f (x) = cn einx ,
∀x, y ∈ V , hx, yi = hy, xi∗ ; linearity bra : hφ| (lin. function), ket: |ψi
j=0
wrt. 2nd argument & positive defi- (vector). |αψ + βχi = α|ψi + β|χi. Particular solution to inhom. eq. n=−∞
(2)
niteness. hαψ + βχ| = α ∗ hψ| + β ∗ hχ|. braket:
polarization identity 4hx, yi = hφ|ψi (inner product), outer pro- n 1
Z π
f (s)e−ins ds.
Zx
2 2 2
duct:P|ψihφ|. Completeness relation: X Wi (s) • s1 , s2 ∈ R: (s1 − s2 ) < N → 2 cn =
[ x + y − x − y ] + i[ x − iy − P yp (x) = yi (x) ds, (1) lin. indep. sol. (s1 − s2 ) ∈ N → π −π
2 ψ = i hφi |ψiφi ⇒ I = i |φi ihφi |. W (s)
x + iy ], norm, parallelogram i=1 sol. y1 , y2 = Cy2 from (2).
2 2 2 Convergence to the mean:
2 ODEs

x + y + x − y = 2(kxk2 + y ), Wi (s) = W (s) ↔ (0, ..., b(x))T i th col. Lin. 2nd order ODEs, BVP
angle, orthogonality. System of 1st order ODEs Dirichlet: y(a) = ya , y(b) = yb
Scalar linear higher order ODEs 0 0 0 0 lim kf (x) − fN (x)k
dy
Hom. problem: characteristic eq. → Neumann: y (a) = ya , y (b) = yb N →∞
2 2
= f(x, y(x)): autonomous if f ≡

C.-Schwarz: hx, yi ≤ kxk2 y . dx
f(y), linear if f = A(x)y + b(x), ho- roots. Inhom. problem: cf. (1). Mix: [ya & yb0 ] or [yb & ya0 ] Zπ !1/2
Hilbert spaces Limiting process: solution w. multi- Robin: α0 y(a) + α1 y 0 (a) = γa , = lim |f (x) − fN (x)|dx
mogeneous if b(x) = 0. Cauchy pro- N →∞
Inner product space which is com- dy plicity 2 ∼ limiting case of 2 distinct β0 y(b)+β1 y 0 (b) = γb , (α0 , α1 ) , (0, 0), −π
plete. x† = (x∗ )T . blem: dx = f(x, y(x)) with y(x0 ) = y0 . roots which merge (l’Hôpital).
∂f
(β0 , β1 ) , (0, 0)
Orthogonal/normal basis ∃ & uniqueness if f & ∂y conti- Scalar lin. 2nd order ODEs Periodic: y(a) = y(b) + y 0 (a) = y 0 (b). Half range Fourier series
orthonormal if hφi , φj iδij ∀i, j. nuous. y 00 + p(x)y 0 + q(x)y = r(x). Obtain 1st f (x) over [0, π]. extend f over
→ projection over basis: xi = Linear 1st order ODEs non trivial solution to hom. prob, Sturm-Liouville [−π, π], forcing even f (x) =
hφi , ψi/hφi , φi i. Gram-Schmidt pro- homogneous solution = vec- 2nd sol.: Eigenvalue problems (homoge- f (−x) ⇒ a = 2 R π f (s) cos (ns)ds
neous lin. 2nd order ODE over n π 0
hψj |χi i tor space of dim. n. inhomo- Zx
ds − R s p(t)dt
cess: ψi = χi − i−1 or odd f (−x) = −f (x) ⇒ bn =
P
j=1 hψ |ψ i ψj
j j geneous problem: solution y2 (x) = y1 (x) e , interval [a, b] w. homogneous BCs).
y12 (s) 2 π
R
Form
π 0 f (s) sin (ns)ds.
P
Linear operators y(x) = n λi yi (x) + yp (x). !
Adjoint A† : ∀ψ, χ ∈ V , hψ, Aχi = → scalar: solution −1 d dy
particular solution Ly = p (x) +q(x)y = λy 2L periodic Fourier Series
hA† ψ, χi. (A† )† = A, (αA + βB)† = Rx Z x Rx
ω(x) dx 0 dx
a(s)ds a(t)dt Zx (3)
α ∗ A† + β ∗ B† , (AB)† = B† A† , (A−1 )† = y(x) = y0 e x0 + b(s)e s ds, r(s)y1 (s)
yp (x) = y2 (x) ds + homogeneous BCs at x = a, b
(A† )−1 . x0 W (y1 , y2 )(s) such that L is self-adjoint wrt. X∞
nπx
Self-adjoint (hermitian): H † = H,
Zx Rb f (x) = cn e i L
r(s)y2 (s) ∗
hf , gi = a ω(x)f (x)g(x)dx, p0 , ω
normal: [A, A† ] = 0. AB self adjoint → system (A ∈ Rn×n constant): − y1 (x) ds. n=−∞
W (y1 , y2 )(s) continuous and p0 (x), ω(x) > 0 for ∞
⇔ [A, B] = 0. Unitary: U −1 = U † & a nπx nπx
X
U ψ = ψ .
Z x x ∈ [a, b]. Problem = finding par- = 0 + an cos +bn sin
y(x) = e(x−x0 )A y0 + e(x−s)A b(s)ds. ticular parameter λ for which ∃ 2 L L
n=1
Matrix representation: y = Ax, Frobenius method
j+s , a , 0. Or- non trivial solution (eigenvalue and
x0
y(x) = ∞
P
if {φi } orthonormal basis, Aij = j=0 aj (x − x0 ) 0 eigenfunctions of L).
hφi , Aφj i. If A = adjoint operator: dinary pt: p(x0 ) & q(x0 ) < ∞, can !!!: L self adjoint ⇔ BT cancel as ZL
eA = PeDA P−1 DA = P−1 AP. be expanded in positive integer se- result of appropriate BCs ⇒ homo- 1 nπs
f (s)e−i L ds
(A† )ij = (A∗ )ji . cn =
ries converging for |x − x0 | < R (∃ 2 geneous BCs and periodic if p0 (x) 2L −L
Basis transformation: x0 = Bx, φj = lin. indep. sol., s1 = 1, s2 = 0). Reg. periodic too are OK. 1 L
Z
Pn 0 0 nπs
i=1 Bij φi . If {φi }, {φi } both ortho- Scalar higher order ODEs P∞ j
singular pt: p(x) = j=−1 pj (x − x0 ) , an = f (s) cos ds
0
normal: Bij = hφi , φj i ⇔ B unitary.   Casting ODE into self-adjoint form L −L L
dn y dn−1 y P∞ j
= φ x, y, ..., ≡ system of q(x) = j=−2 qj (x − x0 ) . 00 0
a2 (x)y + a1 (x)y + a0 (x)y = λy ⇒ (3)
B : φi → φi0 ⇒ A0 = BAB−1 . dxn 1 L
Z
dxn−1 nπs
n 1st. order ODEs, cf. function Inject series into ODE, identify with bn = f (s) sin
Eigenvalue problem: Aφ = λφ, φ ∈ terms w. same powers: lowest = in- x
L −L L
di−1 y
(Z )
V , 0, λ ∈ F. χA (λ) = det(λI − A). Ei- z(x) = (zi (x)), zi (x) = dxi−1 ⇒ dz n
dx =
2
dical eq. I(s) = s + (p−1 − 1)s + q−2 = ω(x) =
C
exp
a1 (s)
ds
genbasis of normal operators: nor- dz(x) 0 → roots s1 ≥ s2 . Higher: I(s+j)aj = a2 (x) a2 (s) a − ibn a + ibn
mal operator on V ⇒ ∃ othonormal φ(x, z1 , ..., zn ) ⇒ dx = f(x, z(x)), w. ⇒ cn = n , c−n = n
eigenbasis of V . Self-adjoint opera- f = (z1 , ..., zn , φ)T .
Pj−1 
− k=0 (k + s)pj−k−1 + qj−k−2 ak →
 p 0 (x) = −a 2 (x)ω(x) 2 2
tor ⇒ λi ∈ R, {φi } ⊥. rec. q(x) = a0 (x).
 MMP cheatsheet - PH EPFL, 2021 Parseval’s identity Can be seen as limit of sequence g(z) = za f (z) avec nb. fini de sin- FT and applications to PDEs
Author: Xavier KERVYN, Page 2 /2 gularités zk ∈ C, aucune sur l’axe Useful for unbounded problems.
∞ n 2 2
réel non-négatif etR tq. |za f (z)| → 0 Convolution enables to fold in the
X δn (x) = √ e−n x (differentiable)
Aperiodic function f over [a, b] hf (s)|g(s)i = 2L gn fn∗ π ∞
(z → ∞, z → 0). ⇒ 0 f (x)xa−1 dx = initial conditions or BCs. Approach:
construct half range series over n=−∞   apply FT on x on both sides of the
2πi P a−1 , z
[−L, L], where L = b − a. 1−e 2πia k Res f (z)z k PDE (not on t), use properties to
Parseval’s theorem sin nx obtain an ODE in k-space with
• cosines: f (z) over [0, L], f (−z) δn (x) = (useful w. FT) FT and inverse FT in 3D independent variable t. Then con-
over [−L, 0], where z = x − a. πx Zn vert inital conditions in FT-space.
1 Finally apply inverse FT on the
→ δn (x) = eiωx dx.
Z
∞ 1
• sines: f (z) over [0, L], −f (−z) X
2π −n f (~r) =
~
e−i k·~rg(~k)d3 k solution ψ̂ found for the ODE to
over [−L, 0], where z = x − a. hf (s) | f (s)i = 2L |cn |2 (2π) 3/2 3
n=−∞ ZR obtain solution in x-space. Use
Integration of Fourier Series Fourier transform 1 ~ convolution and FT of an integral.
g(~k) = ei k·~sf (~s)d3 s
 ∞ 

 |a0 |2 X  Fourier Transform 3/2
= L  + 2 2
|an | + |bn |  (2π) R3
2 5 Useful tricks / Miscellaneous
Z ∞
Z x
a0 L
n=1 1 iωs Nouveaux vecteurs de base:
F [f ] = g(ω) = √ e f (s)ds Elementary properties of FT
f (x)dx = (x − x0 ) 2π −∞ ~ ~k ~
x0 2π ~ = ei R·
• F [f (~r − R)] g(k) ∂~
x/∂r ∂~
x/∂ϕ
∞ Solving PDEs with FS êr = , êϕ = , etc.
X an L nπx (BCs + initial conditions if time). 3 Inverse Fourier Transform x/∂r|
|∂~ x/∂ϕ|
|∂~
+  sin steps method: Z∞ • F [f (α~r)] = |α|−1 g(α −1~k)
nπ L −1 1
n=1
x 1. Obtain the ODES: ψkl = F [g] = √ e−iωx g(ω)dω • F [f (−~r)] = g(−~k) êr = ∇r
= cos ϕ êx + sin ϕ êy ,
∞
bn L nπx  2π −∞ |∇r|
Fk,l (t)Gk (x)Hl (y) → substituti- ∇ϕ
X
+ cos  • F [f ∗ (~r)] = g ∗ (~k) êϕ = |∇ϕ| = − sin ϕ êx + cos ϕ êy ,
nπ L  on & isolation of each varia- ∇z
n=1 x0 ble. êz = |∇z| = êz .
Important Fourier Transforms • F [∇f (~r)] = −i~kg(~k)
ω2
1st convolution thm 2. solve the ODEs satisfying BCs F [e−at 2 ] = √1 e− 4a , Re(a) > 0 Lipschutz: une façon d’obtenir la for-
to obtain Fk,l , Gk and Hl . √ 2a • F [∇2 f (~r)] = −k 2 g(~k)
me diagonale PT AP est d’effectuer
Then write the eigenmodes F [1] = 2πδ(ω) √ une suite d’opérations élémentaires
Z L ψkl (with constants determi- F [δ(x)] = 1/ 2π FT of an integral
(f ∗ g)(x) = g(s)f (x − s)ds sur les lignes et d’effectuer la même
ned from initial conditions). fˆ(ω−a)+fˆ(ω+a) suite d’opérations élémentaires sur
−L F [f (x) cos(ax)] = 2
"Z x #
∞ ˆ ˆ −g(k) les colonnes. Ces deux suites de
X nπx 3. Solve the PDEs satisfying F [f (x) sin(ax)] = f (ω−a)−f (ω+a) F f (s)ds = + πg(0)δ(k) mêmes opérations élémentaires sur
= 2L fn gn ei L
initial conditions: express 2i −∞ ik
I nous donneront PT . Cf. Algèbre
q
n=−∞ the solution as the sum of F [e−a|x| ] = π2 2 a 2 , Re(a) > 0 lin. Schaum, p.264.
a +ω Fourier convolution
all the possible P eigenmo- F [xn ] = i n √2πδ(n) (ω)
2nd convolution thm des ψ(x, y, t) = k,l ψkl (x, y, t). Polynômes caract 3x3 et 2x2:
F [ 1x ] = −i π2 sgn(ω)
p
Apply the initial conditions Z
χ(λ) = λ3 − Tr(M)λ2 + (M1 + M2 +
∞ F [sgn(x)] = π2 iω 1 U (s)V (r − s)ds M3 )λ − det(M) ;
i nπx
X
χ(λ) = λ2 − Tr(M)λ + det(M)
R
h(x) = f (x)g(x) = hn e L , Z
X πkx πly Calculus of residues H
n=−∞ ψ0 (x, y) = Mk,l sin sin Residues theorem: C f (z)dz = = Û (k)V̂ (k)e−ikr dk
∞ Lx Ly R
k,l P
2πi k Res(f , zk ). Identités de somme et différence:
X
hn = gm fn−m πly (3D version OK).
X πkx
m=−∞ ν0 (x, y) = ωk,l Nk,l sin sin
Lx Ly Poles of order n: Res(z0 ) =
sin α + sin β = 2 sin
α+β
cos
α−β
k,l 1 dn−1
| (z − z0 )n f (z). Parseval’s theorem 2 2
Translation property (n−1)! dzn−1 z=z0
α+β α−β
and find coefficients by projection. Fourier: nb. fini de singularités zk
Z Z cos α + cos β = 2 cos cos
∞ dans demi-plan supérieur C \ R et
∗
U (s)W (s)ds = Û (k)Ŵ ∗ (k)dk 2 2
X nπx 4 Fourier Transforms & PDEs α+β α−β
f (x − x0 ) = f˜n ei L , tq. lim|z|→∞ |f (z)| R= 0 si Im(z) ≥ R R
sin α − sin β = 2 cos sin
Dirac delta 2 2
n=−∞ 0. ⇒ ∀p > 0, R f (x)eipx dx = or equivalently
nπx δ(x) = 0, x , 0. α+β α−β
−i L 0
 
f˜n = fn e 2πi Im zk >0 Res f (z)eipz , zk . cos α − cos β = −2 sin sin
P
hW |U i = hŴ |Û i. 2 2
Z ∞
f (0) = f (x)δ(x)dx Mellin: a ∈ C, Re(a) > 0, a < Z+ ,
−∞