TransitLeastSquares.jl

A Julia implementation of the Transit Least Squares algorithm (Hippke & Heller 2019) for detecting periodic transits in photometric time series. Designed to be feature-compatible with the reference Python package transitleastsquares while taking advantage of Julia's multi-threading and SIMD for substantially higher throughput.

Installation

using Pkg
Pkg.add("TransitLeastSquares")   # once registered in the Julia General registry

Quick start

using TransitLeastSquares

# time in days, flux normalized to ~1
result = tls(time, flux; flux_err = flux_err)

result.period        # best-fit period (days)
result.T0            # transit mid-time of the first transit (days)
result.duration      # transit duration (days)
result.depth         # fractional transit depth
result.SDE           # signal detection efficiency
result.power         # detrended power spectrum vs result.periods

Catalog lookup

If you have a TIC / KIC / EPIC identifier, catalog_info fetches stellar parameters and limb-darkening coefficients:

info = catalog_info(TIC = 307210830)
result = tls(time, flux;
             R_star = info.R_star, M_star = info.M_star, u = info.u)

Parallelism

The period search parallelises over the period grid. Launch Julia with the number of threads you want to use:

julia -t auto --project -e 'using TransitLeastSquares; ...'

Benchmarks

Head-to-head against the reference Python transitleastsquares (v1.32) on an identical 5000-point synthetic light curve (injected P=3.5 d, T0=1.2 d, duration=0.15 d, depth=0.01, white noise σ=5e-4, 60-day baseline; grid covers 1.0-8.0 d at oversampling=2, yielding ~2380 trial periods; run on a 28-core Linux box):

Metric	Python TLS (1T)	Python TLS (4T)	TransitLeastSquares.jl (1T)	TransitLeastSquares.jl (4T)
Wall time	10.46 s	4.75 s	5.05 s	1.41 s
Period (true 3.5)	3.500254	3.500254	3.498740	3.498740
T0 (true 1.2)	1.197572	1.197572	1.209165	1.209165
Duration (true 0.15)	0.167294	0.167294	0.153819	0.153819
Depth (true 0.01)	0.010603	0.010603	0.010751	0.010751
SDE	39.485	39.485	33.929	33.929
Transit count	17	17	17	17
Period grid size	2380	2380	2377	2377

At matched thread count TransitLeastSquares.jl is ~3× faster; against Python's default single-thread config it is ~7× faster. All recovered transit parameters agree with Python TLS to within the grid resolution. The residual SDE gap (33.9 vs 39.5) comes from the underlying χ² values — TransitLeastSquares.jl currently uses an analytic small-planet Mandel-Agol template rather than the full batman model Python TLS uses; the SR/power/SDE statistic itself is normalized identically (see src/statistics.jl).

Reproduce with:

julia --project benchmark/generate_lc.jl
python3 benchmark/run_python.py
julia --project -t 4 benchmark/run_julia.jl

Citation

If you use TransitLeastSquares.jl in published work, please cite Hippke & Heller (2019):

@article{Hippke2019,
  author  = {Hippke, Michael and Heller, Ren\'e},
  title   = {{Optimized transit detection algorithm to search for periodic transits of small planets}},
  journal = {Astronomy & Astrophysics},
  volume  = {623},
  pages   = {A39},
  year    = {2019},
  doi     = {10.1051/0004-6361/201834672}
}

Status

Early development (v0.1). See test/runtests.jl for the validated feature set.