Scalar Wave Eqn in 1+1D expanded as envelope and carrier
(∇2+ k20)E = 0 with E =E(x, z)eik0z
∂2E
∂z2 +∂2E
∂x2 + k02= 0 = eik0z
✓✓✓✓
✓
✓
✼0SV EA
∂2E
∂z2+ i2k0
∂E
∂z −❙k❙❙20+∂2E
∂x2+❙k❙❙20
= 0
leads to evolution eqn
i2k0∂E
∂z +∂2E
∂x2 = 0
If we have sampled representation of the transverse field E(x) → Enwe will need to evaluate derivatives along x to advance the field along z.
Which side should we take difference on?right or left?
dE
dz ≡En+1− En
∆x
= δxr dE
dz ≡En− En−1
∆x
= δlx
simple, but not very accurate. Errors build up with the number of numerical integra- tions.
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 464
Numerically Evaluating Derivatives
En+1= E(x + ∆x) = E(x) + E′(x)∆x+2!1E′′(x)∆2x+3!1E′′′(x)∆3x+ . . .
⇒ δrx= E′(x) +∆x
2 E′′(x) + . . . δlx= E′(x)−∆x
2 E′′(x) + . . . Both approximations have error at 2nd order (are thus 1st order accurate).
How about averaging these two estimates δxE =12 δxr+ δlx
E =En+1− En−1
2∆x
= E′(x) +∆2x
3 E′′′(x) + . . . ∼= dE dx Much better: 2nd order accurate!
Second derivative δx2En= δx
En+1− En−1
2∆x
=
En+2−En
2∆x −En−E2∆xn−2
2∆x
=En+2− 2En+ En−2
(2∆x)2 But better to use adjacent points with same 2nd order difference scheme
d2E
dx2 ∼= δlxδxrEn≡En+1− 2En+ En−1
(∆x)2
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 465
Runge-Kutta
Consider an evolution eqn y′= f (x, y)
eg [x → z, y → E] inhomogeneous absorption or 2-photon nonlinear absorption
dE
dz =−α(z)E dE
dz =−β2|E|2E Euler method has O(∆x)error
yn+1= yn+ ∆xyn′ = yn+ ∆xf (xn, yn) For O(∆2x)error use 2nd order Runge-Kutta
yn+1= yn+12[∆xf (xn, yn)
+∆xf (xn+ ∆x, yn+ ∆xf (xn, yn)]
y
x yn
yn+1 y(x)
n∆x (n+1)∆x
y
x yn
yn+1 y(x)
n∆x (n+1)∆x
}
Q Q Q
Q Q Q
Q Q Q
Q Q Q
Q Q Q Q
Q Q Q Q Q Q
Q Q
Q
10−6 10−4 10−2 100
relative error
errors in Runge−Kutta First order
Second order
PDEs
Paraxial Diffraction
∂E
∂z + 1 i2k0
∂2E
∂x2 = 0 Normalize variables x ← xk0
∂
∂zu(x, z) + i∂2
∂x2u(x, z) = 0
Represent sampled field at x = j∆x, z = n∆zas unj. Simple discrete evolution eqn un+1j − unj
∆z
= iunj−1− 2unj + unj+1
(∆x)2 ⇒ un+1j = unj + ∆ziunj−1− 2unj + unj+1 (∆x)2 Canexplicitly advance field in z using known unj. Unfortunately this isunstable.
implicit scheme calculates transverse spatial derivative∂x2∂2 after evolution step in z un+1j − unj
∆ = iun+1j−1 − 2un+1j + un+1j+1 (∆ )2
Tridiagonal System
unj =−
aj
z}|{
i∆∆z2
x un+1j−1+
bj
z }| {
hi2∆∆z2
x− 1i un+1j −
cj
z}|{
i∆∆z2
x un+1j+1
un1 un2 ...
unj−1 unj unj+1
...
unJ
=
b1 c1
a2 b2 c2
... ... ...
aj−1 bj−1 cj−1
aj bj cj
aj+1 bj+1 cj+1
... ... ...
aJ bJ
un+11 un+12 ...
un+1j−1 un+1j un+1j+1 ...
un+1J
soln
start with β1= b1, γ1= un1/b1then iterate to j = 2, ..., J βj= bj−ajcj−1
βj−1
γj=unj − ajγj−1
βj
un+1j = γj−cjun+1j+1 βj
and un+1J = γJ
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 468
Stability Analysis
Consider a plane wave u(x) = eikx≡ eik∆xj= eiκj Solution with z (eg n) will be of the form
u(n, j) = ξneiκj
The stability is determined by |ξ(κ)| > 1 unstable for some κ. For explicit scheme δzru(n, j) = iδ2xu(n, j) = ieiκ(j−1)− 2eiκj+ eiκ(j+1)
(∆x)2 ξn=−iξn eiκj
(∆x)2 e−iκ− 2 + eiκ un+1j − unj =−iξneiκj 4
(∆x)2sin2(κ/2) un+1j = ξneiκj− i∆zξneiκj 4
(∆x)2sin2(κ/2)
= ξneiκj
1− i 4∆z
(∆x)2sin2(κ/2)
| {z }
= ξn+1eiκj
|ξ| > 1 for ALL κ ⇒ unstable For κ ≪ 1 |ξ| ∼= 1and OK accuracy but for κ ∼= 1, get rapid growth
Noise associated with high frequencies ( that we are not interested in) blows up
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 469
Lax method for improving stability
Replace unj →12(unj+1+ unj−1)with smoothed average in longitudinal derivative un+1j − unj
∆z
= iunj−1− 2unj+ unj+1
(∆x)2 ⇒ un+1j −12(unj+1+ unj−1)
∆z
= iunj−1− 2unj + unj+1 (∆x)2
un+1j =12(unj+1+unj−1)+i∆zδ2xunj=12ξn
eiκ(j+1)+eiκ(j−1) +i∆z
eiκ(j−1)−2eiκj+eiκ(j+1) (∆x)2 ξn
=12ξneiκj2 cos κ− i∆z
∆2xeiκjξn4 sin2(κ/2)
Stability Conditioncos κ − i4∆∆z2xsin2(κ/2) ≤ 1 ⇒ cos2κ+ 4∆∆z2
xsin2(κ/2)2
≤ 1
1−κ2
2
+ 42
∆z
∆2x
2κ
✓✓2
4
≤ 1 ⇒ ∆z
∆2x<
√2 4 Why? rewrite
un+1j − unj =12 unj+1− 2unj + unj−1
+ i∆zδ2xunj ⇒ i∂
∂zu +
1− i∆z
∆2x
∂2
∂x2u again for κ ≪ 1 (low freq) |ξ| ∼= 1and good accuracy. Dissipitave term added to eqn for κ ∼= 1 dz|ξ| < 1 and exponential decay (high frequency noise decays away).
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 470
Crank-Nicholson
Why use un+1j after propagation for∂x∂22derivative instead of values of unj before?
How about averaging these two approaches un+1j − unj
∆z
= i 2
"
un+1j−1− 2un+1j + un+1j+1
(∆x)2 +unj−1− 2unj + unj+1 (∆x)2
#
unconditionally stable! second order accurate in both x and z implicit, requires tridiagonal solution
a b c d e f
− z}|{i 2∆2xun+1j−1+
z }| {
i
∆2x+ 1
∆z
un+1j −
z}|{i 2∆2xun+1j+1 =
z}|{i 2∆2xunj−1−
z }| {
i
∆2x− 1
∆z
unj +
z}|{i 2∆2xunj+1
b ca b c
... ... ...
a b c
a b c
a b c
... ... ...
a b
un+11 un+12
...
un+1j−1 un+1j un+1j+1
...
un+1J
=
e f d e f
... ... ...
d e f
d e f
d e f
... ... ...
d e
un1 un2 unj−1...
unj unj+1
u...nJ
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 471
Split-Step
i∂
∂zE + L{E} + f(|E|2)E = 0
Where L{E} is a linear operator (eg ∂x∂22, or nonparaxial diffraction) and f() is an operator describing the inhomogeneities and/or nonlinearities
integration of this type of first order equation results in exponential. In the case of just the linear operator
E(z + ∆z) = ei∆zLE(z) = ei∆z∂x2∂2E(z) =X
m
i∆z∂2
∂x2
m
m! E(z)
While just in the presence of nonlinearity/inhomogeneity E(z + ∆z) = ei∆zf (∆z/2)E(z)
In the presence of both
E(z + ∆z) ∼= ei∆zLei∆zf (z)E(z) = ei∆zf (z+∆z)ei∆zLE(z)
= ei∆zL/2ei∆zf (z+∆z/2)ei∆zL/2E(z)
Symmetrized split-step approach gives 2nd order accurate result
∆z 2∆z
2
∆z Evaluate inhomogeneity in middle of step
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 472
Beam Propagation, sampling, and Fourier Space
Suppose we know a sampled version of our field at some transvers plane z = 0, and wish to propagate it to subsequent planes z = m∆z, while obeying the differential equations containing terms such as diffraction, index inhomogeneities and lenses, ab- sorption variations, and other inhomogeneous and nonlinear effects.
∆x=2λ
∆x=λ
∆x=λ/2
∆x
∆z E(n∆x)
x
∆k= 2π Ν∆x
∆k= 2π Ν∆x
∆k= 2π Ν∆x
Real Space
Fourier
Space k-space
Advance Sampled Field
2-wavelengths
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 473
Beam Propagation: Paraxial vs Nonparaxial
Given input on planar boundary, find FT
E(kx, ky) =Fxy{E(x, y)}
Each transverse component propagates with its own phase factor kz=q
k02− kx2− k2y= k0
s
1−kx2+ ky2
k20 =≈ k0−kx2+ k2y 2k0
Transfer function (phase only for propagating kx, kysuch that kx2+ k2y≤ k20) Hz(kx, ky) = e−ikz(kx,ky)z= e−iz√
k02−k2x−ky2≈ e−ik0zei
k2x+k2y 2k0 z
Fourier spectrum at a distance z
E(kx, ky; z) =E(kx, ky; 0)Hz(kx, ky)
k-space version of transfer function vs Feit and Fleck
Phase rate of the transfer function eikzz= ei√
n2k20−kx2−k2yzwith z can be written as kz=
r n2ω2
c2 +∇2⊥=q
n2k02− kx2− k2y= nk0+q
n2k20− k2x− k2y− nk0
Where the second form separates out the fast varying and slower varying part, but when used over very wide angular ranges would be superflous.
Feit and Fleck use a less intutive version when introducing Fourier beam propagation
eiβz= e
iz
"
k+ ∇2⊥
k+√k2+
∇2T
#
= e
iz
"
nk0− k2x+k2y nk0+√
n2k20−k2x−ky2
#
Equivalence can be shown by multiplying thru by the denominator q
n2k20− kx2− ky2
h
nk0+q
n2k20− k2x− ky2
i
= nk0
h
nk0+q
n2k20− k2x− k2y
i−kx2+ky2
nk0
qn2k20− k2x− ky2+ nk02− kx2− k2y= n2k20+ nk0
qn2k02− k2x− k2y− k2x− ky2
Beam Propagation: Sampling requirement
Consider homogeneous rectangular region of length L and width W (we will do 1-D analysis).
Transversely sample the field at a spacing of ∆x(∆y) being sure to obey the Nyquist condition for a field of bandwidth B.
∆x≤ 1 2Bx
# samples N = W
∆x
This gives a sampled transverse field En= E(n∆x).
Now divide up the longitudinal propagation distance L into M steps of spacing ∆z. How big can we choose ∆z– As big as we want!
(only true for homogeneous case, and can result in periodic wrap around) Typically make phase factor < π for highest frequency component
B2x 2k0
∆z< π say ∆z< 2πk0
10Bx2 or increase spacing ∆zuntil you notice change of the final result.
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 476
Beam Propagation Algorithm
Propagate sample field in Fourier domain
FFT {E(n)} = F (n; 0)
F (n; 1) = F (n∆k; 1· ∆z) = F (n; 0)H1(n)
∆k=2πW =N ∆2π
Transfer function over a distance 1 · ∆x z
H1(n) = e−i√
k20−(n∆k)2∆z
≈ e−ik0∆zein2(∆k)
2 2k0 ∆z
Invert transform to get field back in real space.
IFFT {F (n : 1)} = E(n∆x; 1· ∆z) Repeat to step through all M slices sequentially.
E(n;m) FFT
F(n;m)
F(n;m+1) E(n;m+1)
IFFT
Repeat Propagate
H (n)1
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 477
Beam Jumping: Only for homogeneous media and without boundary absorptionb
However each time you go around this loop you can accumulate numerical noise Why? isnt digital computing perfect? Roundoff errors
How could you minimize or alleviate the accumulation of such numerical noise? Use double precision complex.
E(n;0) FFT F(n;0)
H (n)1
H (n)2
H (n)3
H (n)4
H (n)m
IFFT E(n;1) IFFT E(n;2) IFFT E(n;3) IFFT E(n;4) IFFT E(n;m) log RMS[E(n;m)-E (n;m)]actual
Avoid any accumulated noise by beam jumping.
FT input field F (n; 0) = FFT {E(n; 0)}
multiply by transfer function to get directly to m∆z
Hm(n) = e−i√
k20−(n∆k)2m∆z
Inverse transform and repeat for all m
E(n; m) =IFFT {F (n; 0)Hm(n)}
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 478
Inhomogeneous Beamprop
Inhomogeneous and nonlinear index effects n = n0+ δn(~r, E) k2=
π λ0
2
(n20+ 2nδn + δn2)≈ k20(1 + 2δn/n0) Start with scalar wave eqation
∂2E
∂z2 +∂2E
∂x2 + k2E = 0 Expand E as envelope and carrier E(x, z) = a(x, z)e−ik0z Insert into wave eqn and use SVEA∂∂z2a2 ≈ 0
∂2E
∂z2 = e−ik0z
∂2a
∂z2− 2ik0∂a
∂z− k20a
e−ik0z
−2ik0∂a
∂z− k02a +∂2a
∂x2+ k20(1 + 2δn/n0)a
= 0 Evolution equation
∂a
∂z = −i 2k0
∂2a
∂x2+ k202δn n0
a
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 479
Inhomogeneous Beamprop
Formal solution: integrate in z starting with B.C. E(x, 0) = a(x, 0) a(x, z) = e−2k0i
´z 0
h∂2a
∂x2+k022δnn0ai
dza(x, 0) Seperate into Index step followed by diffraction step
Index step in real space
small step in z, δz – define average inhomogeneous index δn = 1
∆z
ˆ z+∆z z
δn(x, z)dz 2nd term of operator is multiplication by complex phase factor
N∆za(x, z) = e−i2πλ0δn(x,z)∆za(x, z) Diffraction step in Fourier space
FT envelope in transverse x coordinate a(x, z) =
ˆ
A(kx; z)eikxxdkx
A(kx; z) = ˆ
a(x, z)e−ikxxdx
E(x, z) = ˆ
E(kx; z)eikxxdkx
E(kx; z) = ˆ
E(x, z)e−ikxxdx
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 480
Inhomogeneous Beamprop
Partial derivative easily evaluated as multiplication in Fourier space∂x∂ ⇐⇒ (ikx)
∂2a
∂x2 ⇐⇒ (ikx)2A = −kx2A A(kx; z+∆z) =A(kx; z)e−i(√
k20−kx2−k0)∆z
≈ A(kx; z)ei2k0k2x∆z ⇐⇒ M∆za(x, z) E(kx; z+∆z) =E(kx; z)e−i√
k02−k2x∆z≈ E(kx; z)e−ik0∆zei2k0k2x∆z ⇐⇒ M∆zE(x, z) Complete Propagation step – only 1st order accurate (M∆zN∆z: Why not N∆zM∆z?)
a(x, z + ∆z) =Fx−1
ne−ikz(kx)∆zFx
a(x, z)e−ik0δn∆z o
=M∆zN∆za(x, z) Symmetrized split-step operator – 2nd order accurate
a(x, z + ∆z) = M∆z2 N∆zM∆z2 a(x, z)
= Fx−1
(
e−ikz(kx)∆z2 Fx
e−ik0δn∆zFx−1
ne−ikz(kx)∆z2Fx
a(x, z) o)
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 481
Rectangular aperture diffraction
dark=bright aperture
dark=dim obstruction
Solid:
Exact RS Dotted:
Paraxial Dashed:
BPM Formal Derivation
Evolution Equation to be solved in the form of operator Q = L+N consisting of linear and inhomogeneous parts
∂A
∂z = iQA Formal solution
A(z) = ei´0zQ(z′)dz′A(0)
over a small propagation distance ∆zthe linear part, which is constant can be integrated A(∆z) = ei
h´∆z
0 L(z′)dz′+´∆z 0 N (z′)dz′i
= ei
h
∆zL+´∆z 0 N (z′)dz′i
A(0) Approximate integral over the spatially varying nonlinear/inhomogeneous part ˆ ∆z
0 N (z′)dz′= ˆ ∆z
0
N
∆z
2
+
z′−∆z
2
∂
∂zN
∆z
2
+O(∆2z)
dz′
=N
∆z
2
∆z+1 2
z′−∆z
2
2
∆z
∂
∂zN + O(∆3z) =N
∆z
2
∆z+O(∆3z)
Symmetrical Split Step
Approximation to integral of nonlinear operator is 2nd order accurate, but we dont know self consistent field at∆2z. BPM approximates this field after linear step of ∆2z neglecting NL contribution. Results in symmetrized split step that is 2nd order accurate in ∆z
A(∆z) = ei∆z2Lei∆zN (∆z/2)ei∆z2LA(0)
Where the linear propagation is most easily applied in the Fourier domain ei∆z2LA(0) =F−1n
ei∆z2√
k20−k2TF{A(0)}o
Note when applying these steps, we can coalesce adjacent linear half steps into a full step unless we need to know field at full step locations.
A(2∆z) = ei∆z2Lei∆zN (∆z/2)e|i∆z2L{zei∆z2L} ei∆zN (∆z/2)ei∆z2LA(0)
= ei∆z2Lei∆zN (∆z/2)ei∆zLei∆zN (∆z/2)ei∆z2LA(0)
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 484
Numerical Evaluation of BPM Error
ǫ =
pP||Atest| − |Aref||2
|Aref|2
Comparison versus minimum step size Comparison versus theoretical solution
Shows second order accuracy up to large steps where it becomes first order because of the innacuracy of the estimation of the field at the half step. Roll-off at small ∆zis due to discretization in the transverse dimension with step size ∆x.
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 485
2+1D Beam propagation
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 486
Circular Aperture 3+1D Beam Propagation Crossection
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 487
Imaging with Fresnel Zone Plate
✓z✓✓o2+ h2= (zo+ δ)2=✓z✓✓o2+ 2zoδ + ✒0
δ2 ✓z✓✓2i + h2= (zi+ δ′)2=✓z✓✓i2+ 2ziδ′+ ✒0 δ′2 δ = h2
2zo
δ′= h2 2zi
∆ = δ + δ′= h2 2zo
+ h2 2zi
is OPD Successive zones with an aditional half wavelength OPD are
labeled as successive fresnel zones with radial boundaries hm
∆m=mλ 2 =h2m
2
1 zo
+ 1 zi
⇒ hm= smλ
1/f =p mλf Area of mth annulus bounded by hm−1and hm
Am= πh2m− πh2m−1= π(mλf− (m − 1)λf) = πλf circular apertures that consist of N zones will sum on-axis fields out of phase with equal amplitude contributions
h1 h2 h3 h4 h5 h6 h7 h8
zo zi
hm δ δ’
AT OT = A1− A2+ A3− A4+· · · ± AN =
(N odd ≈ A1⇒ IT OT= A21 N even ≈ 0 ⇒ IT OT = 0
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 488
2-D crosssections every 8 λ from a
D = 16λ Circular Aperture BPM
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 489
Circular disk diffraction: Fresnel/Arago’s
Bright Spot Babinet’s principle and quadrature
combination with incident plane wave
u0
ua
u0 = +u ua d
u0
ua
ud = −
No absorbing boundary conditions:
wraparound can interfere producing unwanted fringes
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 492
Absorbing boundary conditions: Smoothly taper to avoid reflection at impedance
discontinuities
Circulant wraparound due to FFT computation Abrupt absorbing edge causes reflection due to impedance mismatch
Absorbing Boundary condition to avoid wraparound, smooth to avoid reflection
xi=IndGen(nx)-(nx/2) abwdth=nx/128 boundabsx=(1-exp(-(nx/2-abs(xi+.5))/abwdth)) xphase=EXP((-j*2*!PI*delz*(conj(SQRT(DCOMPLEX((n0)ˆ2-((xind)/xsize)ˆ2))) )))
FOR z = 1, nz-1 DO BEGIN field = FFT(FFT(field,-1)*xphase,1)*boundabsx ENDFOR
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 493
Absorbing boundary conditions: power may not be conserved
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 494
Beam Propagation in 1-D without Additional Aberrations
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 495
Beam Propagation in 1-D with 1 wave Spherical Aberration
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 496
Beam Propagation in 1-D with 2 wave Spherical Aberration
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 497
Beamprop through Lens Systems
Double slit diffraction and Fourier Transform Comparison of BPM with theory
Thick Lenses
Thick Singlet Lens with Aberations showing spherical aberration
Multiple Incoherent Input Beam showing field curvature, 1D coma
BeamPropagation through 4F lens system
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 500
BeamPropagation through 4F lens system
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 501
4F lens system with Schlieren filter Converts Phase Modulation to Amplitude
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 502
4F lens system with Schlieren filter Converts Phase Modulation to Amplitude
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 503
4F lens system with Zernike Phase contrast Converts Phase Modulation to Amplitude dot
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 504
4F lens system with Zernike Phase contrast Converts Phase Modulation to Amplitude dot
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 505
Beamprop in coupled singlemode waveguides
Waveguide separation .5 λ Waveguide Coupled Modes
Waveguide separation 2 λ Waveguide Coupled Modes
Beamprop in wide multimode waveguides
Waveguide separation 0 λ Waveguide Coupled Modes
Wide multimode waveguide Waveguide Coupled Modes
Arc in k-space from Fourier transform of BPM of Gaussian Beam
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 508
2+1D Beam Propagation of Nonparaxial Spot array: Aberration of Free Space
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 509
spot array through focus showing off-axis aberrations of free space
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 510
3D Fourier space of nonparaxial spot array
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 511
k-space propagation
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 512
k-space propagation
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 513
Projection of 3-D k-sphere onto kx− k
z plane to get FT of x-z crosssection of circular
aperture
Comparing k-space (optics centric) vs
the Ewald sphere (Material centric)
for X-ray discrete lattice diffraction
k-space and the McCutchen Theorem
s
P
O W
Q
f
q R
k-sphere
2π λ
kz kx
Aperture
R
k
Axial Converging Spherical wave Cut out by aperture
s− f = ˆq · ~Rfor P near geometric focus dS = f2dΩ≈ s2dΩ
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 516
McCutchen Theorem
C.W. McCutchen, Generalized Aperture and the Three-Dimensional Diffraction Image, JOSA Vol 54, pg 240, 1960.
U ( ~R) =Ae−ikf iλf
¨
W
eiksi
s dS = A iλ
¨
ΩW
e−ikˆq· ~RdΩ = 1 iλ
˚
e−ik ~Q· ~RdVk
ˆ
qis a unit vector over all angles
Q~ is not a unit vector, so must enforce A(Q) = Q(q)δ(|Q| − 1).
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 517
McCutchen Theorem
C.W. McCutchen, Generalized Aperture and the Three-Dimensional Diffraction Image, JOSA Vol 54, pg 240, 1960.
3D Annular aperture is 2D projected onto the z-axis as a rectangular projection so 1-D slice through z-axis in region of focus is sinc
Gaussian beam illuminating aperture has one sided exponential projection so z-axis slice is Lorentzian
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 518
Wide Angular diffraction and Fourier space projection- slice : .4
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 519
Wide Angular diffraction and Fourier space projection- slice : .512
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 520
Wide Angular diffraction and Fourier space projection- slice : .8
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 521
Wide Angular diffraction and Fourier space
projection- slice : .96 Comparison of conventional BPM with FT
of circular arc in Fourier space
Apodization of the aperture
Min Gu, Advanced Optical Imaging Theory, Springer, (2000) ch 6
r
f = g(θ)is ray projection function
dS = 2πrdr = 2πf2g(θ)g′(θ)dθand dΩ = 2πf2sin θdθ p2(r)dS = P2(θ)dΩ Conservation of energy p2(r)2πrdr = 2πf2g(θ)g′(θ)dθ = P2(θ)2πf2sin θdθ P (θ) = p(r)g(θ)gsin θ′(θ)
Sine Condition Ray in image space meets the focal sphere at same height at which corresponding ray in object space enters system
g(θ) = sin θ r = f sin θ P (θ) = p(r)√ cos θ Herschel Condition Ray density constant over wavefront
g(θ) = 2 sinθ
2 r = 2f sinθ
2 P (θ) = p(r)
r ray density p(r)
P(θ) θ Angular ray density
Fresnel Transmisson
dS
dΩ
Uniform Projection Condition Equal radial distances converted to equal angular in- tervals on the reference surface
g(θ) = θ r = f θ P (θ) = p(r)
r θ
sin θ Helmholtz Condition Distortion free imaging
g(θ) = tan θ r = f tan θ P (θ) = p(r) cos−3/2θ
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 524
Dispersion in Lenses and femtosecond pulses
Propagation Time Difference Group Velocity Dispersion
Can analyze polychromatic case by coherently summing up contributions from each spectral component, appropriately weighted by complex amplitude spectra.
This works for both incoherent sources and femtosecond laser pulses Alternatively can analyze the spatio-temporal pulse envelope evolution
Use differential equations and analytically solve for pulse evolution
Use 2+1D spatio-temporal beam propagation with dispersion like diffraction where reduced time in group velocity frame T = t − z/vgis transverse dimension
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 525
Fourier representation of Green’s function
A. Ba˜nos Jr, Dipole Radiation in the presence of a conducting half-space, Pergamon, 1966
Scalar waves expand according to a spherical wave Green’s function g(R) =eikR
R
Where the Green’s fnc satisfies Helmholtz eqn driven by point source (∇2+ k2)g =−4πδ(x)δ(y)δ(z)
Now represent this Green’s function in Fourier space G(kx, ky, kz) =
˚ ∞
−∞
g(x, y, z)e−i(kxx+kyy+kzz)dx dy dz g(x, y, z) = 1
(2π)3
˚ ∞
−∞
G(kx, ky, kz)ei(kxx+kyy+kzz)dx dy dz
where G(kx, ky, kz) represents an analytic function of the 3 transform variables kx, ky, kz.
To compute G(kx, ky, kz) multiply both sides of inhomogeneous Helmholtz eqn by e−i(kxx+kyy+kzz)and integrate wrt x, y, z
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 526
3-D k-space
˚
e−i(kxx+kyy+kzz)(∇2+ k2)g(x, y, z)dx dy dz
=
˚
e−i(kxx+kyy+kzz)
∂2g
∂z2+∂2g
∂y2+∂2g
∂x2
dx dy dz + k2G(kx, ky, kz)
=
˚
e−i(kxx+kyy+kzz)[−4πδ(x)δ(y)δ(z)] dx dy dz = −4π Derivative theorem turns derivatives to (ikj)so this becomes
(−kx2− k2y− k2z+ k2)G(kx, ky, kz) =−4π So the 3-D Fourier transform of the scalar Green’s function is
G(kx, ky, kz) = 4π kx2+ k2y+ k2z− k2 In 2-D this would be
G(kx, kz) = 4π k2x+ kz2− k2
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 527
Spherical Coordinates
Transforming spherically symmetric functions
x = R sin θ cos φ kx=K sin α cos β y = R sin θ sin φ ky=K sin α sin β
z = R cos θ kz=K cos α
kxx+kyy+kzz = R sin θ cos φK sin α cos β + R sin θ sin φK sin α sin β + R cos θK cos α
= RK cos α By rotating coordinates to align R with z ⇒ cos θ = 1 This spherically symmetric 3-D FT of the 3-D Green’s function can be represented as
G(K) = 4π K2− k2
Inverse transform must be independent of θ, φ can be represented as g(R) = 4π
(2π)3 ˆ ∞
0
ˆ π 0
ˆ 2π 0
eiKR cos α
K2− k2K2dK sinα dα dβ =2π 2π2
ˆ ∞
0
K2 K2−k2
ˆ π 0
eiKR cos αsinα dα dK
= 1 π
ˆ ∞
0
K✓✓2 K2− k2
eiKR cos α
−i✚KR✚✚
π
0
dK = 1
−iπR ˆ ∞
0
e−iKR− eiKR
K2−k2 KdK K′= K
= 1 iπR
ˆ ∞
−∞
eiKR
K2− k2KdK =eikR R
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 528
2-D K-sphere as complex variable with pole
G(kx, kz) = 4π kx2+ k2z− k2
Kelvin Wagner, University of Colorado Fourier Optics Fall 2019 529
K-space Beam Propagation R.K. Kupka JOSA v12(2) p. 404 1995
2-D (x,z) propagation of eiωtsideband results in TE and TM scalar equations
∇2+ k20n2(x, z)Ey(x, z) = 0
∇2+ 1 n2(x, z)
∂n2(x, z)
∂x
∂
∂x+ k02n2(x, z)
Hy(x, z) = 0 Both can be solved using BPM with matrix representation of operators.
Discretize fields across width W by m samples of spacing ∆x= W/m.
Ey(x)→ ~E(x) = [E(−m∆x/2), . . . , E(0), . . . , E((m/2− 1)∆x)]
FFT of field array gives k-space vector, E(kx)with spaing ∆k= 2π/W multiplication by a function replaced by a m × m matrix
n(x)→ diag[n(−m∆x/2), . . . , n(0), . . . , n((m/2− 1)∆x)] n(x) = diag[n(x)]
Operators represented by matrices, eg Fourier transform
K-space Beam Propagation
k matrix k → k = diag[−m∆k/2, . . . , 0, . . . , (m/2− 1)∆k]allows Fourier transform of real space differentiation
∂ ~E(x)
∂x =−iF−1kE(kx) dual Convolution theorem F
n2(x)
∗ F{E(x)} = F
n2(x)E(x)
becomes NE(kx) = F n2(x)· ~E(x)
where N defined by the commutation n2(x)F−1= F−1N is equivalent to a convolu- tion N → n2(kx)∗ Column vectors of N contain FT of n2(x)but they are shifted along the mian diagonal as banded diagonal Toeplitz matrix
F F T{n2(x)} = [a−∆km/2, . . . , a−∆k, a0, a∆k, . . . , a∆k(m−1)/2]
... ... ... ... ... . . . . . . a0 a−1 a−2 a−3 a−4 . . .