After some literature search, we choose to start with the AMBRE stellar library, which covers 300-1200nm at R~150,000.  The parameters covered are 2500 K ≤ T eff ≤ 8000 K, −0.5 ≤ log(g) ≤ 5.5 dex, and
−5.0 ≤ [M/H] ≤ +1.0 dex, which should be sufficient for our purpose.  The wavelength coverage is slightly short and we need to extend the library to 1300nm.  Also, the parameter sampling is somewhat coarse and we need to interpolate between the templates.

We experimented several functional forms to extrapolate to 1300nm.  A simple exponential form works very well.

The blackbody (lambda^(-4)) does not work very well contrary to our expectations.

What should the simulator insert above 1200nm? The flat extrapolation?

Either model fit extrapolation won't include Pa-beta at 1282nm. The PFS web page says the instrument only goes out to 1260nm, so this is OK, right?

We do not need to care about Paschen beta as you say. F-stars do not have any strong absorption lines at 1200-1300nm. There are weak absorption lines and we probably need to mask them when we derive the flux calibration vector. We could also ask the author of the AMBRE library to extend the wavelength coverage. I was in touch with the author and he told me that he could extend the library to 1300nm if he had a line list.
cloomis I am sorry, but I am missing the context here. Do you want to use this library for the 2d simulator? We plan to generate an extended set of spectra with interpolated parameters and with extrapolation to 1300nm.

Masayuki Tanaka: The extension of the AMBRE library to 1260nm is now the subject of a more specific issue, PIPE2D-364. I believe the current issue is to choose a stellar library. So given that you choose AMBRE, I propose that we close this issue on that basis.

But before we close this issue, please add a comment to this ticket explaining why AMBRE was chosen.

Apologies for having left this ticket open.  The reasons why we chose AMBRE are
1 - It covers a wide wavelength range, extending into the nearIR.  Most stellar libraries in the literature cover only the optical wavelengths, but AMBRE extends out to 1.2um.
2 - It has a sufficiently high spectral resolution. We do not want to use libraries with lower resolution than PFS.

3 - It covers a wide range of stellar parameters.

It is a theoretical library, but we confirmed that the location of the AMBRE spectra on a color-color diagram is consistent with that of observed stars from HSC.