SI Appendix
SI Appendix
SI Appendix
Junxiao Zhou, Haoliang Qian, Ching-Fu Chen, Junxiang Zhao, Guangru Li, Qianyi Wu,
Hailu Luo*, Shuangchun Wen, and Zhaowei Liu*
Supplementary text
Figs. S1 to S4
References for SI reference citations
1
www.pnas.org/cgi/doi/10.1073/pnas.1820636116
Supplementary Information Text
Detailed formulations
This designed phase gradient metasurface is based on the Pancharatnam-Berry (PB phase), which
is related to polarization change and is created by the local orientation of the optical axis of the
metasurface (1,2). As the orientation of the metasurface is locally variant, the metasurface can be
characterized by a position-dependent Jones formalism (3). The Jones matrix of metasurface with
spatially varying local optical axes and constant phase retardance 𝜋 can be given as: (4)
cos 2𝜑 sin 2𝜑
𝑇0 = [ ] (1)
sin 2𝜑 −cos 2𝜑
𝜋𝑥
Here, 𝜑(𝑥, 𝑦) = Λ
is normally the orientation of local optical axis. Given the fact that sample is
located at the Fourier plane, the field distribution is calculated by the Fourier transform of the
RCP object light
𝐸̃𝑖𝑛,𝐹𝑜𝑢𝑟𝑖𝑒𝑟 (𝑢, 𝑣) = ℱ[𝐸𝑖𝑛,𝑜𝑏𝑗 (𝑥, 𝑦)] |𝑅⟩ (2)
Where 𝑢 = 𝑥 ′ /𝜆𝑓, and 𝑣 = 𝑦 ′ /𝜆𝑓, which 𝑥 ′ and 𝑦 ′ is the real space coordinate at Fourier plane.
𝜆 is the working wavelength and 𝑓 is the focus length. |𝑅⟩ = (1, 𝑖)𝑇 and ℱ is the Fourier-
transform operator. The transmitted field after the sample can be determined by
1
𝐸̃𝑜𝑢𝑡,𝐹𝑜𝑢𝑟𝑖𝑒𝑟 (𝑢, 𝑣) = 𝑇0 𝐸̃𝑖𝑛,𝐹𝑜𝑢𝑟𝑖𝑒𝑟 (𝑢, 𝑣) = 𝐸̃𝑖𝑛,𝐹𝑜𝑢𝑟𝑖𝑒𝑟 (𝑢, 𝑣) exp (i2𝜑) [ ] (3)
−𝑖
As we can see, this process brings in an additional phase 2𝜑, which is the PB phase in nature
(Fig1. (c)). Meanwhile, the handedness of output beam is reversed. The field in image plane
could be written as:
𝐸𝑜𝑢𝑡,𝑖𝑚𝑔 (𝑥, 𝑦) = ℱ[𝐸̃𝑜𝑢𝑡,𝐹𝑜𝑢𝑟𝑖𝑒𝑟 (𝑢, 𝑣)]
1
= ℱ{ℱ[𝐸𝑖𝑛,𝑜𝑏𝑗 (𝑥, 𝑦)]} ∗ 𝛿(𝑥 − Δ) [ ]
−𝑖
1
= 𝐸𝑖𝑛,𝑜𝑏𝑗 [(𝑥 − Δ), 𝑦] [ ] (4)
−𝑖
𝜆𝑓
where ∗ represents the convolution operation and the image shift Δ = Λ
, as shown in Fig1. (e).
Similarly, the input electric field of one LCP incident, |𝐿⟩ = (1, −𝑖)𝑇 , the achieved PB phase
will be written as −2𝜑. The sign ‘−’ is due to PB phase spin-dependent property. And the output
of our system can be given as
1
𝐸𝑜𝑢𝑡_𝑖𝑚𝑔_𝐿𝐶𝑃 (𝑥, 𝑦) = 𝐸𝑖𝑛_𝑜𝑏𝑗_𝐿𝐶𝑃 [(𝑥 + Δ), 𝑦] [ ] (5)
𝑖
It is well known that a linearly polarized (LP) beam can be regarded as the combination of LCP
and RCP components. Therefore, for attaining spatial differentiation, we let a LP beam incidents
2
the metasurface, two shifted images along the opposite direction are obtained at the image plane,
which could be given as:
1 1
𝐸𝑜𝑢𝑡_𝑖𝑚𝑔_𝐿𝑃 (𝑥, 𝑦) = 𝐸𝑖𝑛_𝑜𝑏𝑗_𝐿𝑃 [(𝑥 − Δ), 𝑦] [ ] + 𝐸𝑖𝑛_𝑜𝑏𝑗_𝐿𝑃 [(𝑥 + Δ), 𝑦] [ ] (6)
−𝑖 𝑖
Here, the analyzer is set after the sample, which is orthogonal with the input linear polarizers. The
output electrical field could be rewritten as:
0
𝐸𝑜𝑢𝑡_𝑒𝑑𝑔𝑒 (𝑥, 𝑦) = (𝐸𝑖𝑛_𝑜𝑏𝑗_𝐿𝑃 [(x + Δ), y] − 𝐸𝑖𝑛_𝑜𝑏𝑗_𝐿𝑃 [(x − Δ), y]) [ ] (7)
𝑖
Thus, the amplitude of the output electric field can be further given as, |𝐸𝑜𝑢𝑡_𝑒𝑑𝑔𝑒 (𝑥, 𝑦)| =
(𝐸𝑖𝑛_𝑜𝑏𝑗_𝐿𝑃 [(x + Δ), y] − 𝐸𝑖𝑛_𝑜𝑏𝑗_𝐿𝑃 [(x − Δ), y]). If the shift Δ is much smaller than the image
profile, the output electric field amplitude will be approximately proportional to the first-order
spatial differentiation of the input electric field:
d𝐸𝑖𝑛 (𝑥, 𝑦)
|𝐸𝑜𝑢𝑡_𝑒𝑑𝑔𝑒 (𝑥, 𝑦)| ≃ 2 Δ d(𝑥)
(8)
Figs. S1 to S4
3
Fig. S1. Objects for edge detection.
Fig. S2. Experiment setup. L, lens. P, Glan laser polarizer. CCD, charge coupled camera. An experiment
setup built to validate the edge detection generated by our sample is shown in Figure S2. The fundamental
Gaussian beam comes from a supercontinuum lase (SuperK EXW-12, NKT Photonics). The incident light
then passes through the object, which is followed by a 4f system. The sample (metasurface) is placed in
between of the P1 and P2.
Here, we experimentally measured the transfer function. As shown in Figure S3(a), the
laser beam passes through a 10 mm lens and then incidents to the spatial differentiator.
The splitting beam spot is collected by the second focus lens and then collected by the
CCD camera. The distance between the first lens and metasurface, metasurface and the
4
second lens, and the second lens and CCD camera are equal to the focal distance, 10 mm.
For measuring the transfer function H(𝑘𝑥 ), we first measure the intensity distribution
𝐼1 (𝑥, 𝑦) at the location showing in Fig. S3 (c) and calculate the electrical field distribution
based on 𝐸1 (𝑥, 𝑦) = √ 𝐼1 (𝑥, 𝑦). And then move the CCD camera before the L1 (Fig. S3
(b)), record the intensity distribution 𝐼0 (𝑥, 𝑦), and get the electrical field distribution
𝐸 (𝑢,𝑣)
𝐸0 (𝑥, 𝑦). The final transfer function H(𝑘𝑥 ) = 𝐸1 (𝑢,𝑣) . Here 𝑢 = 𝑥/𝜆𝑓, and 𝑣 = 𝑦/𝜆𝑓 ,
0
which 𝑥 and 𝑦 is the real space coordinate at Fourier plane. 𝜆 is the working wavelength
and 𝑓 is the focus length. Note that our spatial differentiator works on one direction, so
we set y=0 and the transfer function is demonstrated on the x direction.
Fig. S3. Measurement of the spatial spectral transfer function on the proposed spatial differentiator (red
dashed box). (a) Experiment setup: Laser (He-Ne laser, 632.8nm wavelength); L1, L2, lens with focal
length 10 mm; GLP1 and GLP2, a pair of crossed Glan laser polarizer; Metasurface, period 1000 μm;
CCD, charge couple device. (b)- (c) The experimental intensity distribution before and after passing
through the whole system, respectively. (d) The experiment results of the transfer function.
5
Fig. S4. Simulation result of two-dimensional edge detection. (a) The image of object. (b) The simulation
results of one-dimensional edge detection. (c) Simulation result of two-dimensional edge detection.