Skip to content

smooth_gaussian

Smooth a spectrum by convolution with a Gaussian kernel.
Category Smoothing
Backend scipy - implemented here on top of scipy primitives
Version 2.0.0
Reads ctx.spectrum (a Spectrum1D)
Writes spectrum

Non-finite samples (NaN/inf) are bridged by linear interpolation over the finite neighbours before the convolution, so a single bad pixel no longer spreads NaN across the kernel footprint ; the original non-finite positions are restored to NaN in the output (finite inputs are processed bit-identically to v1.0.0). v2.0.0 — uncertainty propagation : when the input carries an uncertainty, σ_out,i = sqrt(Σ_k w_k² σ_{i+k}²) with the very kernel the filter applies (normalised Gaussian truncated at 4σ) ; for white noise of σ and sigma = 2 px this gives σ_out ≈ 0.376 σ (≈ σ / sqrt(2 σ_px √π)) in the interior. Within int(4σ + 0.5) samples of either end the filter's 'nearest' padding replicates the end sample, so the kernel taps falling off the array add before squaring — the propagation reproduces the filter's exact linear operator there too, and the edge σ_out is larger (0.641 σ at the very first sample for sigma = 2). Bridged (non-finite) samples contribute no variance and get a NaN uncertainty. The correlation the kernel introduces between neighbouring output samples is NOT tracked — χ² fits on a smoothed spectrum under-estimate their parameter errors. Without an input uncertainty the brick is unchanged from v1.0.1.

Parameters

Parameter Default Required Description
sigma 2.0 - Standard deviation of the Gaussian kernel, in samples.

Use it

from spectro_kernel import run_algorithm

output = run_algorithm("smooth_gaussian", ctx, {
    "sigma": 2.0
})

spectro run smooth_gaussian --input spectrum.fits \
  --param sigma=2.0

{
  "tool": "smooth_gaussian",
  "arguments": {
    "session_id": "<session_id>",
    "params": {
      "sigma": 2.0
    }
  }
}

Every algorithm is an MCP tool of the same name; describe_algorithm returns this page's metadata as JSON.

References

  • scipy.ndimage.gaussian_filter1d
  • Bevington & Robinson 2003, Data Reduction and Error Analysis for the Physical Sciences, 3rd ed., McGraw-Hill — §3.2, error propagation through a linear combination (σ_out² = Σ w_k² σ_k²).