Running AO simulations with scintillation (doFresnel = True)

[LucaInaf]

I have been running AO simulations that include scintillation, i.e., using physical propagation by setting doFresnel = True in the prop object of the params_pyr_soul.yaml file in SOUL. I did not modify the gain or the ff.

I have noticed that when I display the pupil phase, the values are typically smaller compared to the same AO setup without scintillation (around 400 peak-to-valley versus 4000 or more). Additionally, the Strehl ratio does not seem to vary consistently when changing the seeing conditions, for example from 1.5 arcsec to 0.9 arcsec. It fluctuates around 75%, and sometimes it drops significantly before returning to higher values.

Is this behavior expected? In general, does it make sense to include scintillation effects for 8-meter-class telescopes?

[guidoagapito]

Hi, thanks for reporting this.

Yes, this behavior is completely expected due to the mathematical and numerical nature of physical wave propagation.

Here is a breakdown of what is happening:

Why the pupil phase values are smaller:
When you use geometric propagation (doFresnel = False), the simulation linearly adds the phase screens, resulting in a continuous, "unwrapped" phase that can span thousands of nanometers.
When doFresnel = True, the phase is calculated by extracting the angle from the propagated complex electric field.
This strictly confines (wraps) the phase between -pi and +pi, which translates to roughly +/- lambda/2.
For visible wavelengths, your peak-to-valley phase will physically never exceed a few hundreds of nm.

Why the Strehl ratio fluctuates and drops:
Physical propagation using the Angular Spectrum method relies heavily on Fast Fourier Transforms (FFTs).
If the padding_factor is not manually set to a high enough value, the FFTs will suffer from cyclic convolution (aliasing). Diffracted light leaving the edges of the grid will wrap around to the other side, severely corrupting the wavefront and causing the erratic Strehl behavior you observed.
Additionally, feeding heavily wrapped phase maps into a standard AO control loop can cause instabilities.

Does it make sense for 8m-class telescopes? Generally, no.
The spatial scale of atmospheric intensity fluctuations (scintillation) is defined by the Fresnel zone (r_F = sqrt(lambda L), lambda is the propagation wavelength L is layer altitude), which is typically around 5 to 10 cm.
Over an 8-meter aperture, these intensity variations are heavily smoothed out due to a phenomenon called aperture averaging. Unless you are simulating a High-Contrast Imaging (Extreme AO) coronagraph system where amplitude aberrations severely limit the contrast, scintillation is negligible.

Wavelength Dependence (Chromaticity):
Unlike geometric propagation, physical Fresnel propagation is highly chromatic.
The diffraction patterns and resulting amplitude variations depend strictly on the wavelength (the wave vector k = 2 pi lambda is hardcoded into the propagator).
If your simulation involves a Wavefront Sensor (WFS) and a Science instrument operating at different wavelengths (e.g., Visible vs. Near-IR), a single physical propagation at a reference wavelength will not yield accurate results for both.
You would technically need separate propagation setups for each wavelength, which adds significant complexity and computational overhead.

Because physical propagation requires careful tuning of numerical parameters (like padding) to avoid artifacts, and since we currently lack comprehensive documentation/tutorials for it, we strongly advise keeping doFresnel = False for standard AO simulations.

[LucaInaf]

Thank you very much for the explanation.
Can I instead run open-loop simulations safely, or should this be done with caution?

[guidoagapito]

Yes, running open-loop simulations is actually a very good idea and a much safer way to start experimenting with this feature.

Starting in open-loop will allow you to safely build confidence with physical propagation.

• Observe the actual intensity fluctuation patterns at the pupil plane and how they change with seeing conditions.
• Test different values for the padding factor to see exactly how it suppresses numerical artifacts.
• Get a feel for the computational cost of the Angular Spectrum method compared to geometric propagation.
• The physical and numerical constraints still apply. Even in open-loop, if the padding factor is too low or if you ignore the wavelength dependence, your resulting wavefronts and Point Spread Functions (PSFs) will still be corrupted by numerical issue.