Xray Scattering from Random Rough Surfaces
Abstract
This paper presents a new method to model Xray scattering on random rough surfaces. It combines the approaches we presented in two previous papers – PZ&LVS[zhao03] & PZ[zhao15]. An actual rough surface is (incompletely) described by its Power Spectral Density (PSD). For a given PSD, model surfaces with the same roughness as the actual surface are constructed by preserving the PSD amplitudes and assigning a random phase to each spectral component. Rays representing the incident wave are reflected from the model surface and projected onto a flat plane, which is the first order approximation of the model surface, as outgoing rays and corrected for phase delays. The projected outgoing rays are then corrected for wave densities and redistributed onto an uniform grid where the model surface is constructed. The scattering is then calculated using the Fourier Transform of the resulting distribution. This method provides the exact solutions for scattering in all directions, without small angle approximation. It is generally applicable to any wave scatterings on random rough surfaces and is not limited to small scattering angles. Examples are given for the Chandra Xray Observatory optics. This method is also useful for the future generation Xray astronomy missions.
Further author information: Ping Zhao: Email: zhao@cfa.harvard.edu
1 Introduction
The study of wave scattering from rough surfaces goes back at least to Rayleigh, in his 1877 classic – The Theory of Sound[rayleigh], which led to the development of the Rayleigh criterion (see Section 2) for classifying the degree of surface roughness. Since then, the problem of scattering from random rough surfaces has been investigated by many physicists and engineers, and has been the subject of many books, including the classic – The Scattering of Electromagnetic Waves From Rough Surfaces by Beckmann and Spizzichino[beckmann] and countless research papers[ogilvy]. A good solution for Xrays scattering in grazing incidence on random rough surfaces is in high demand due to the emerging fields of Xray optics in many applications. This problem is even more difficult due to the short wavelength (comparing to the scale of the surface roughness) and the small angle between the wave propagating direction and the surface. Most approaches in the literature make the approximation that the scattering angle is much smaller than the incident grazing angle. Some of the treatments use the approximation that the surfaces are sufficiently “smooth” so that a low order expansion in the surface height errors is adequate, and consequently are limited in their applications. Most of those methods can not obtain the scattering asymmetry around the direction of specular reflection (scattering towards versus away from the surface). These approximations are not adequate for many of the applications involving Xray mirrors.
In 2002, we presented a SPIE paper – A new method to model Xray scattering from random rough surfaces (PZ&LVS)[zhao03] which introduces a novel approach to the problem and gives the solution for scattering in the incident plane. In 2015, we presented a second SPIE paper – Transverse Xray scattering on random rough surfaces (PZ)[zhao15] which gives the solution of scattering in the direction perpendicular to the incident plane. Based on the above two SPIE papers, this paper provides a complete and exact solution for Xray scattering from random rough surfaces in all directions. This new method is generally applicable and provides the exact solution to all wave scattering problems on random rough surfaces.
In previous methods, scattered wave was usually treated separately as coherently reflected wave in the specular direction and incoherently scattered (or diffused) wave in other directions (see e.g. Simonsen[simonsen]). However, for any given wave, there is no clear distinction between “smooth” and “rough” surfaces (see Section 2), therefore it is impossible to separate scattered wave into coherently reflected and incoherently scattered waves.
The novelty of our new method is that it treats the reflected wave and scattered wave together (e.g. every ray is treated as scattered, even in the specular direction) as coherent scattering, and consequently both depend upon the surface roughness. It does not require any assumptions in order to separate the wave into reflected and scattered waves, and does not require the small angle approximation so that all the scattered rays can be traced accurately.
This new study of the century old problem is motivated by our direct involvement of the evaluation of the Xray mirror performance aboard the Chandra Xray Observatory (CXO) – the NASA’s third great space observatories, which has been successfully operated since July 23, 1999. It is the first, and so far the only, Xray telescope achieving subarcsec angular resolution ( FWHM), which let us see the Xray Universe we had never seen before. Chandra’s spectacular success owes to the genius design and superb manufacture of its Xray mirrors. These mirrors are the largest and the most precise grazing incidence optics ever built. At 0.84m long and 0.6 – 1.2m in diameters, the surface area of each mirror ranging from 1.6 to 3.2 square meters. They were polished to the highest quality ever achieved for any Xray mirrors of this size. The surface roughness of these mirrors is comparable to or less than the Xray wavelengths in the 0.1–10 keV band over most of the mirror surfaces. However, the mirrors are still not perfect, and consequently there are still small amount of scattered Xrays. We need an accurate model of the Xray scattering to fully evaluate the Point Spread Function (PSF) of the CXO in order to correctly understand the scientific data. In this paper, we use the Chandra mirror surface roughness data as examples to illustrate the application of this new scattering method. This method is also useful for the future generation Xray observatories.
2 Random Rough Surfaces
A rough surface is a surface that deviates from the designed or assumed perfect surface upon which an incident wave achieves perfect specular reflection without energy loss to any other directions. In reality, there is no absolutely perfect surface exist. Almost all the surfaces, from the rough ocean to mountainous land, from the deck of an aircraft carrier to the finest polished mirrors, either natural or man made surfaces, are rough surfaces.
Also, no two rough surfaces are identical even they were both formed in the same wellcontrolled process. Not only that, every part of the same surface is also unique. We can not predict the exact roughness on one part of the surface from our knowledge of the other parts of the same surface. Its profile is simply “random”. These kind of surfaces are called random rough surfaces. Statistical methods are required to describe and study them, such as the Power Spectral Density (PSD, see Section 3). Given the PSD of a particular surface, we still can not describe the exact roughness of that surface, but we can describe a series of surfaces with the same roughness, which reflect and scatter incident waves the same way as the original surface.
When study wave reflection and scattering, a surface is called “smooth” if it specularly reflects the energy of an incident plane wave into one direction; whereas a surface is called “rough” if it scatters the energy into various directions. Based on this definition, the same surface can be called smooth or rough depends on the wavelength and incident angle. Rayleigh first studied the sound wave scattering from rough surfaces in 1877 that led to the development of the Rayleigh criterion[rayleigh]. Consider two parallel adjacent rays, with wavelength , incident on a rough surface with surface height difference , at a grazing angle . Upon reflection in the specular direction, the path difference between the two rays is
(1) 
therefore the phase difference is
(2) 
When the surface is perfectly “smooth”, hence ; two rays are in phase, therefore reflect specularly. When , two rays cancel each other, there is no specular reflection; hence the surface is called “rough”. In this case, the energy are scattered into other directions due to the energy conservation. By arbitrarily choosing the value halfway between these two extreme cases, , we obtain the Rayleigh criterion for smooth surface:
(3) 
In reality, there is no clear cut between the socalled “smooth” and “rough”. A surface satisfying the Rayleigh criterion can only be considered as “nearly” smooth. Unless , there is always a small amount of energy being scattered into nonspecular directions. As we shall see, the Chandra telescope mirrors do satisfy the Rayleigh criterion. But just that small amount of energy scattered away from the specular direction needs to be taken into account in order to fully understand its PSF. In this sense, all the surfaces can be considered as “rough”, because there is no real surfaces are perfect (). Table 1 gives some examples of “smooth” surface based on the Rayleigh criterion.
Surface  Wave frequency/energy  

Airport radar dish  3 GHz  10 cm  80  1.3 cm 
Satellite TV dish  10 GHz  30 mm  60  4.3 mm 
Mirror in your bath room  545 THz  550 nm  90  69 nm 
Chandra Xray Observatory  0.1 – 10 keV  1.24Å – 124Å  27.1 – 51.3  10Å 
3 Power Spectral Density of Rough Surfaces
A rough surface is described, statistically, by its surface Power Spectral Density (PSD) as a function of the surface spatial frequency . Consider a 1dimensional surface with length and surface height (i.e. deviation from a perfectly flat surface): . Its PSD is defined as:^{1}^{1}1The definition is conventional, where the subscript denotes 1dimensional; the PSD satisfies , and typically positive frequency limits are used for most spectral integrals. The total power, , is the integral of from to , i.e. .
(4) 
The PSD, as it is defined, is the “spectrum” of the surface roughness. Its value at is simply the “power” at that frequency. It is easy to distinguish between periodic and random rough surfaces from their PSDs. For periodic rough surfaces, there are some “spectral lines” in their PSDs; whereas there are no lines exist for a real random rough surface.
Given a PSD function , the surface roughness amplitude RMS in the frequency band of – (both and are positive) can be calculated as:
(5) 
For a 2D random rough surface, the two orthogonal dimensions can be treated separately and each as a 1D surface. So the above definitions are still valid. For example, for the grazing incident rays upon a 2D random rough surface, one dimension can be chosen as the intersection of the incident plane and the surface. The scattered rays will remain in the incident plane. This case is called the inplane scattering. The second dimension can be chosen as the direction on the surface but perpendicular to the incident plane. In this case the scattered rays are out of the incident plane. This case is called the outplane, or transverse, scattering. Since the surface roughness can be different in these two directions, their PSDs can also be different.
4 Chandra Xray optics
The Chandra Xray optics – High Resolution Mirror Assembly (HRMA) – is an assembly of four nested Wolter TypeI (paraboloid and hyperboloid) grazing incidence mirrors made of Zerodur and coated with iridium (Ir)[lvs97, zhao97, zhao98, zhao04]. The eight mirrors are named P1,3,4,6 (paraboloid) and H1,3,4,6 (hyperboloid), due to historical reasons (there were 6 mirror pairs when the HRMA was designed and two pairs were later removed to reduce the cost). The mirror elements were polished by the then Hughes Danbury Optical Systems, Inc. (HDOS). The surface roughness was measured during the HDOS metrology measurements after the final polishing, but before the iridium coating[reid95]. Tests conducted on sample flats before and after the coating indicate that the coating does not change the surface roughness.
The instruments used for the measurements were the Circularity and Inner Diameter Station (CIDS), the Precision Metrology Station (PMS), and the Micro Phase Measuring Interferometer (MPMI, aka WYKO). The CIDS was used to determine the circularity and the inner diameters. The PMS was used to measure along individual axial meridians. With these two instruments, HDOS essentially measured the ‘hoops’ and ‘staves’ of each mirror barrel, and thus mapped the entire surface. The microroughness was sampled along meridian at different azimuths using the WYKO instrument at three different magnifications (1.5, 10 & 40)[reid95, zhao95].
These metrology data were Fourier transformed and filtered. The low frequency parts of the CIDS and PMS data were used to form mirror surface deformation (from the designed mirror surface) maps. The high frequency parts of the PMS data and the WYKO data were used to estimate the surface microroughness. Both of them are parts of the HRMA model we built for the raytrace simulation of the Chandra performance.
The mirror surface microroughness has little variation with azimuth, but tends to become worse near the mirror ends. Table 2
HRMA  Sections  Num of  

Mirror  Surface Roughness Amplitude RMS (Å)  Sections  
P1  LC  LB  LA  M (88%)  SA  SB  SC  7  
50.3  8.49  4.51  3.58  4.91  5.94  53.9  
P3  LB  LA  M (92%)  SA  SB  5  
5.37  5.26  1.96  2.38  4.83  
P4  LB  LA  M (93%)  SA  SB  5  
6.41  3.15  2.57  3.21  6.81  
P6  LB  LA  M (94%)  SA  SB  5  
37.1  5.23  3.34  5.65  20.9  
H1  LD  LC  LB  LA  M (88%)  SA  SB  SC  SD  SE  SF  11 
26.9  5.34  3.64  3.34  3.32  3.32  3.32  3.32  3.53  7.30  60.3  
H3  LC  LB  LA  M (92%)  SA  SB  SC  SD  8  
4.87  2.90  2.23  2.08  2.08  2.10  3.95  5.56  
H4  LD  LC  LB  LA  M (93%)  SA  SB  SC  SD  SE  10  
7.18  3.83  2.61  2.57  2.36  2.36  2.74  2.68  4.01  29.4  
H6  LD  LC  LB  LA  M (94%)  SA  SB  SC  SD  SE  10  
19.0  4.92  2.51  2.23  1.95  1.95  1.95  2.07  2.96  15.9  
Total  61 
shows the surface roughness of the 61 HRMA mirror sections based on their roughness. The number underneath each section name is the surface roughness amplitude RMS, , calculated according to Eq. (5) for mm. Each mirror is 838.2 mm in length. The middle sections (M), which are the best polished and hence have the lowest PSDs, cover most part of the mirror surface (the number in parentheses after each M denote the percentage coverage). The ’s for the M sections are only 1.9–3.6 Å. The end sections, where the ’s are relatively higher, cover a very small part of the mirror (), and hence contribute very little to the mirror performance.
It is seen that all the middle sections are polished to the satisfaction of the Rayleigh criterion as “smooth” surface (Eq. 3 and Table 1), therefore they provide very good reflections in the specular direction for Xrays.
However, since the mirror surfaces are not perfect, there are still small among of scatterings from the middle sections. In addition, the end sections are not “smooth” based on the Rayleigh criterion. So we need to have a good scattering model in order to understand the PSF of the telescope.
Figures 2 and 2 show the PSDs of the M (middle) and SC (small end) sections of P1. P1 and H1 are the first polished mirror pair and are slightly “rougher” than other pairs (see Table 2). The (colored) dash and dotted lines show the data from different measurements: the PMS data are in the low frequency range ( mm); the WYKO data with 3 magnifications are in the higher frequency range ( mm. The black solid lines are the combined PSDs from all the measurements. The SC section obviously is much rougher than the M section.
5 Model Surfaces
A random rough surface can be described by its PSD. Most of the existing methods calculate the scattering from the surface PSD. However, our method calculates the scattering directly from the surface profile. Therefore, we first need to construct a model surface that is based on its PSD. From a random rough surface profile, we can derive a unique PSD. But from this unique PSD, we can not reconstruct the original surface, because the phase information was lost when deriving the PSD. However, we can construct any number of model surfaces with the same roughness as the original one from its PSD by assigning different random phase factors to the spectral components.
As mentioned in Section 3, for a 2D surface, we can treat the two orthogonal dimensions separately and each as a 1D surface. To construct a 1D model surface with length , we need to derive consecutive surface heights, , with a fixed interval to cover the surface (i.e. , where is the spatial resolution of the model surface), and their surface tangents, . Appendix A shows that and can be computed from the surface PSD using the following Fourier transforms:
(6)  
(7) 
where is the surface spatial frequency; , , is the assigned random phase factor. Since both and are real, this requires , i.e. and .
To construct the model surfaces of HRMA, we choose and mm. So mm, and mm. Figures 4 and 4 show one set of model surface sections P1M and P1SC, constructed using their PSDs (Fig. 2 and 2) with Eqs. (6) and (7).
Each HRMA mirror is a 2D surface. For any given point on the surface, its two orthogonal dimensions are the one along the meridian and the one along the azimuth. The roughness along the meridian causes the inplane scattering. The roughness along the azimuth causes the outplane scattering. Depending on the polishing method, the roughness can be different in these two directions. For the HRMA, however, the metrology data did not show this difference. Therefore we treat these two dimensions as having the same roughness, i.e. the same PSD is used to construct the model surface profiles in both directions.
6 Inplane scattering from model surfaces
In this section, we derive the inplane scattering formulae of plane incident waves from a model random rough surface. Details of the derivations can be found in Appendices B and C.
We assume the surfaces are sufficiently smooth so that: 1) there is no shadowing of one part of the surface by another; and 2) there is no reflection from one part of the surface to another, i.e. there are no multiple reflections by the same surface. For an incident plane wave with grazing angle , the first condition requires that the absolute values of all the surface tangents, , are less than . The second condition requires less than (when , the reflected wave is parallel to the surface). The first condition is automatically satisfied when the second condition is met. So the surface roughness condition for applying this method is:
(8) 
This condition is easily satisfied for all 61 HRMA sections, as can be seen by comparing the tangent distributions in Figures 4 and 4 with the mean grazing angles of the four HRMA mirror pairs ().
The scattering formula is given by Eq. (74) in Appendix C.4 as the discrete Fourier transform of the field , on axis, in flat surface :
(9) 
where the scattering field intensity is a function of the scattering angle , which is the deviation from the specular reflection direction ( is towards the surface, is aways from the surface); is the wavelength; is the field amplitude, after the reflection, at uniform grid on axis, where the model surface was constructed. is a function of the incident wave, the model surface height and tangent, and the local reflectivity. is a normalization factor given by Eq (79). Again we choose to cover the entire length of the model surface.
Figures 5 and 6 show the inplane scattering results, using Eq (9), for 1.49 keV Xrays incident upon the mirror P1 at its mean grazing angle (51.26). The top two panels show the scattering intensity versus the scattering angle . Positive is defined as the scattering towards the surface; negative is the scattering away from the surface. The sharp peak of specular reflection (topleft) and the Fraunhofer diffraction pattern (topright) are shown as expected. The bottom two panels show the fractional Encircled Energies and the scattering function , defined as:
(10)  
(11)  
(12)  
where , and are the total incident and scattered energy, and the reflectivity of the rough surface as defined in Appendix C.5. The scattering function is the integral of the scattering intensity over the angular space above xaxis. It is simply the probability function (in the domain of [0,1]) of the inplane scattering angle .
7 Outplane scattering from model surfaces
When the scattering angle is small comparing to the incident grazing angle, the transverse scattering angle, i.e. the outplane scattering, is smaller than the inplane scattering angle by approximately a factor of the grazing angle. Therefore traditionally the transverse scattering was treated by simply multiplying the inplane scattering by a factor of the grazing angle in radians. This is a good approximation for small angle scatterings. However, since our method is not limited to small angle scatterings, the above approximation is no longer valid when the scattering angle approaches the grazing angle. In this section, we derive the exact equations for the transverse scattering. Details of the derivation can be found in Appendix D.
The outplane scattering formula is given by the discrete Fourier transform of the field , on axis, in flat surface , as shown in Eq (105) in Appendix D.4:
(14) 
where the scattering intensity is a function of the transverse scattering angle , which is the deviation from the specular reflection direction; is the wavelength; is the field amplitude, after the reflection, at uniform grid on axis, where the model surface was constructed. is a function of the incident wave, the model surface height and tangent, and the local reflectivity. is a normalization factor given by Eq (110).
Figures 7 and 8 show the outplane scattering results, using Eq (14), for 1.49 keV Xrays incident upon the mirror P1 at its mean grazing angle (51.26). The top two panels show the transverse scattering intensity versus the scattering angle . The sharp peak of specular reflection (topleft) and the Fraunhofer diffraction pattern (topright) are shown as expected. The bottom two panels show the fractional Encircled Energies and the transverse scattering function defined as:
(15)  
(16)  
(17)  
where , and are the total incident and scattered energy, and the reflectivity of the rough surface as defined in Appendix D.5. The scattering function is the integral of the scattering intensity over the angular space above yaxis. It is simply the probability function (in the domain of [0,1]) of the outplane scattering angle .
Now let’s compare Fig. 7 of the outplane scattering with Fig. 5 of the inplane scattering, for the mirror section P1M. The topleft panels show the scattering field intensity versus the scattering angle: the outplane scattering is symmetric wrt to the specular direction; while the inplane scattering is asymmetric. This difference is as expected due to the scattering geometry. The other three panels show the core of the intensity, encircled energy and scattering function versus the scattering angle, with the scale of the Xaxis in Fig. 7 smaller than that of in Fig. 5 by a factor of the grazing angle (51.26 = 0.01491 rad). It is seen that these three panels are very similar to each other in the two figures. This indicates that, when the surface is sufficiently “smooth”, therefore the scattering angle is small, the outplane scattering angle is approximately smaller than the inplane scattering angle by a factor of the grazing angle, as shown in Appendix D.6.
Next let’s compare Fig. 8 with Fig. 6, for the mirror section P1SC. The topleft panels still show the symmetric versus asymmetric profile of the outplane and inplane scattering. But, because the P1SC section is much rougher and therefore scattering angle is larger, the other three panels look different, even the scale of the Xaxis in Fig. 8 is still smaller than that of in Fig. 6 by a factor of 0.01491. This indicates that, when the surface is “rough” so that the scattering angles are large, the small angle approximation treatment for the outplane scattering is no longer valid. An exact solution of of the outplane scattering, independent of the inplane scattering, is required for the general case.
8 General solution of scattering from random rough surfaces
With the inplane and outplane scattering formulae Eqs (9) and (14), and their scattering functions Eqs (6) and (7), the general solution of scattering from random rough surfaces can be obtained.
The solution can easily be applied in any raytrace or Monte Carlo simulations. For a monotonic plane wave with a fixed incident angle ,^{2}^{2}2 can be any size and is not limited to grazing angle, because this method works for any incident angle. For normal incident, . In that case, any two orthogonal directions on the surface can be considered as the inplane and outplane; and the two scattering formulae and are identical. if the surface is perfect, the plane wave is reflected in the specular direction, with the reflecting angle . If the surface is not perfect, then every ray is treated as a scattered ray, even in the specular direction.
First, two scattering tables are generated using the inplane and outplane scattering functions and , for the given photon energy and incident angle . Then for each ray, two independent uniform random numbers selected in the domain of [0,1] are used to find the inplane and outplane scattering angles and from the tabulated and (interpolation is needed in this process). Finally the reflected ray is deflected from its specular direction by (inplane) and (outplane), respectively. The result of this raytrace simulation yields a scattering pattern in a 2D angular space for the given plane wave.
However, the above process is only good for one fixed energy and incident angle. For a real source with an energy spectrum and a range of incident angles (e.g. photons hit HRMA mirrors at slightly different angles on the same paraboloid or hyperboloid surface), the above process needs to be expanded to more general cases.
Each pair of scattering functions and are good only for a given photon energy and incident angle . So theoretically, to cover an energy spectrum with a range of incident angles, many and are required on a 2D grid of [, ]. Obviously, this requires enormous amount of computation time and makes the calculation very slow and cumbersome. However, it can be shown that, to a very high degree of accuracy, for small variations of and , and only depend on the product of and , instead of depending on them separately.^{3}^{3}3This can be proven by generating a series of and with different and but keep the product at a fixed value, both and almost stay the same. This fact greatly reduces the amount of computations. Instead of on 2D, and only need to be generated on a 1D grid of .
Define . Since
(19) 
where is the speed of light; is the photon frequency; is the Planck constant. The scattering intensity and (Eqs (9) and (14)) on the 1D grid of can be written as:
(20)  
(21) 
where and are the field amplitudes uniformly distributed on and axis (with even spacing and ) in flat surface , for in and outplane scatterings respectively. They are functions of local surface height, tangent, reflection coefficient, as well as the photon energy and incident angle , as explicitly expressed in Eq (69) in Appendix C.3 and Eq (100) in Appendix D.3.
The scattering functions and (Eqs (6) and (7)) on the 1D grid of can be written as:
(22)  
(23)  
where and are the in and outplane reflectivities as defined by Eq (78) in Appendix C.5 and Eq (109) in Appendix D.5.
The general solution of scattering of plane wave in the photon energy range [], (), and incident angle range [], (), can be obtained in following steps:

For each incident ray with energy and incident angle , use two independent uniform random numbers selected in the domain of [0,1] and interpolation to find its inplane and outplane scattering angles and from the scattering tables.

Finally, the ray is deflected from its specular direction by (inplane) and (outplane), respectively.
The above process yields a scattering pattern in the 2D angular space for the plane wave with a finite range of photon energies and incident angles. This completes the general solution of wave scattering from random rough surfaces.
9 Summary
The exact solution of wave scattering from random rough surfaces are derived. This solution provides a new method to solve the long standing problem of scattering from random rough surfaces in an accurate and more general way. This new method treats both the reflected wave and scattered wave together as coherent scattering, instead of treating them separately as coherent reflection and incoherent (diffuse) scattering. Table 3 compares different aspects of the scattering treatment between the traditional method and the this new method.
Aspect  Traditional Method  New Method 

Scatter and reflection  treated separately as coherent  treated together as 
reflection and diffuse scattering  coherent scattering  
Scattered rays  only some rays are treated as scattered  every ray is treated as scattered 
Scattering angle  much smaller than the grazing angle  no restrictions 
Inplane scattering  symmetric wrt specular direction  asymmetric wrt specular direction 
Outplane scattering  grazing angle times inplane scattering  solved independently 
The major advantage of this new method is that it is not limited by the small angle approximation and gives accurate solutions to inplane and outplane scatterings of any angular size. This made it generally applicable in many wave scattering problems. It is especially useful for Xray scattering at grazing angles. This new method will be very useful for the future Xray astronomy missions.
Appendix A Construction of Model Surfaces
a.1 Fourier Transform
The Continuous Fourier Transform equations are[[cf.]press]:
(24)  
(25) 
Here if is a function of position, , in mm, will be a function of spatial frequency, , in mm.
When there are consecutive sampled values at with the sampling interval , we make the transform:
(26)  
(27) 
where . We obtain the Discrete Fourier Transform equations:
(28)  
(29) 
a.2 Surface Height
From Eq (4), we obtain:
(30) 
Here is a real continuous function of the spatial frequency . We first need to convert Eq (30) to a discrete Fourier transform. Using the equations in A.1 and relation , we obtain:
(31) 
Therefore can be expressed as the forward Fourier transform of as
(32) 
Hence the surface height, , can be expressed as the inverse Fourier transform of
(33) 
where is a random phase factor. A set of surface heights, , can be generated from a set of phase factor . Therefore for a given PSD, we can generate as many sets of surface map (of the same roughness) as we want by changing the random phase factor . Because , the surface height, has to be real, this requires , i.e. and .
a.3 Surface Tangent
Since
(34) 
The surface tangent can be obtained by taking the derivative on both sides of Eq. (34) with respect to :
(35) 
The surface tangent also has to be real. This condition is automatically satisfied because
(36) 
Appendix B Kirchhoff Solution
The wave scattering from random rough surfaces is described by the Kirchhoff solution[beckmann] and its farfield approximation.
As shown in Figure 9, define:

— 2dimensional flat surface at .

— 2dimensional rough surface, described by its surface height .

— incident plane wave (in the incident plane, therefore ).

— reflected or scattered wave from the rough surface .

, — incident and reflecting grazing angles with respect to the surface .
where and are the wave vectors of the incident and scattered waves, so , and
(37) 
A vector normal to the local surface on is given by:
(38) 
The field at an observation point is given by the integration of contributions from the field on the surface :
(39) 
where is an element of surface area; E(s) is given by the incident wave multiplied by the suitable reflection coefficient; the vector goes from the point of integration to the observation point , and ; is a unit vector in the direction of , and . Eq. (39) is known as the general Kirchhoff solution for the wave scattering.
Next we derive the farfield approximation of this solution. When the reflecting surface is near the origin of the coordinate system and the observation point is far from the origin, i.e. when , we have:
(40)  
(41)  
(42) 
where . Keep the first order of in the phase factor and zeroth order elsewhere:
(43)  
(44) 
Eq. (39) becomes:
(45)  
(46)  
(47) 
This is the farfield approximation of the Kirchhoff solution for the wave scattering.
Appendix C Inplane Scattering formula
In this section, we derive the inplane scattering formulae from the Kirchhoff solution for the constructed model surfaces.
c.1 Integral on 1dimensional flat surface
We first isolate the problem by reducing the Kirchhoff solution, Eq (45), to a 1D integral on axis, in flat surface . Figure 10 shows the inplane scattering geometry. Consider:

For scattering in the incident (xz) plane:

For 1D surface in x direction, i.e. only depends on :
(48)  
(49) 
here we have omitted a dimensionless factor , where is the surface length along the yaxis; this factor will be absorbed later in an overall normalization factor .