astro-mmdc

Python SDK for the MMDC astrophysics platform — programmatic access to multi-wavelength SED data, on-demand Swift UVOT/XRT analysis and blazar broadband emission modeling.

MMDC provides APIs for querying astrophysical databases, preparing Spectral Energy Distribution (SED) data from multiple catalogs, running Swift photometry on demand, and running physics simulations for blazar emission modeling using SSC, EIC, and hadronic models.

Installation

pip install astro-mmdc

Or with uv:

uv pip install astro-mmdc

Requires Python 3.10+.

---

Quick Start

Get SED data for a source in 4 lines:

from astro_mmdc import MMDC

client = MMDC()
job = client.sed.prepare_and_wait(ra=187.28, dec=2.05, database_name="3C273")
data = client.sed.get_data(job.uuid)

Run a quick SSC model inference:

result = client.modeling.infer(
    z=0.158, ebl=True, model_type="SSC",
    parameters={"log_B": -1.5, "log_electron_luminosity": 44.0,
                "log_gamma_cut": 5.0, "log_gamma_min": 2.0,
                "log_radius": 16.0, "lorentz_factor": 20.0,
                "spectral_index": 2.2},
)
print(result.nu)     # Frequencies
print(result.nuFnu)  # Fluxes

Run Swift UVOT photometry for a source and time window:

job = client.madam.analyze(ra=166.11, dec=38.21, mjd_start=58849, mjd_end=59031)
for row in job.results:
    if not row.is_lightcurve:
        print(row.obsid, row.filter_band, row.frequency, row.flux)

Or fit a model to data (async batch job):

result = client.modeling.batch_infer(
    "observations.csv",
    z=0.158,
    ebl=True,
    model_type="SSC",
)
print(result.pdf_link)
print(result.best_parameters)

End-to-end pipeline — from sky coordinates to model fit:

from astro_mmdc import MMDC

client = MMDC()

# Fetch SED data
job = client.sed.prepare_and_wait(
    ra=166.11, dec=38.21, database_name="Mkn421", source_name="Mkn 421"
)
info = client.sed.get_info(job.uuid)
client.sed.download_csv(job.uuid, "mkn421_sed.csv")

# Fit SSC model
result = client.modeling.batch_infer(
    "mkn421_sed.csv",
    z=info.redshift,
    ebl=True,
    model_type="SSC",
)

for name, param in result.best_parameters.items():
    print(f"{name}: {param.value:.4f} +/- {param.error:.4f}")

---

Guide

Creating a Client

from astro_mmdc import MMDC

client = MMDC()

# Custom request timeout (seconds)
client = MMDC(timeout=60.0)

# As a context manager (auto-closes HTTP connection)
with MMDC() as client:
    ...

The client provides four resource namespaces:

• client.sed — SED data preparation and retrieval
• client.modeling — blazar emission modeling (SSC, EIC, hadronic)
• client.observations — unified observations catalog (SED + lightcurve)
• client.madam — on-demand Swift UVOT/XRT analysis (MADAM pipeline)

---

SED Data Preparation

SED preparation fetches multi-wavelength observational data from external catalogs for a given sky position.

job = client.sed.prepare_and_wait(
    ra=187.28,              # Right Ascension in degrees
    dec=2.05,               # Declination in degrees
    database_name="3C273",  # Database identifier
    source_name="3C 273",   # Optional display name
)
print(job.status)  # "done", "no_data", or "error"
print(job.uuid)    # Use this UUID for all subsequent data retrieval

If data already exists for a sky position, MMDC returns the cached result. Use force=True to re-fetch:

job = client.sed.prepare_and_wait(ra=187.28, dec=2.05, database_name="3C273", force=True)

Retrieving SED Data

Once a job is complete, retrieve the frequency/flux data:

data = client.sed.get_data(job.uuid)

# data.uuid, data.ra, data.dec, data.source_name
# data.data — nested dict of frequency ranges, catalogs, and flux measurements

Filter by time range, catalogs, or convert units:

data = client.sed.get_data(
    job.uuid,
    mjd_start=50000.0,                      # Filter by MJD time range
    mjd_end=60000.0,
    exclude_catalogs=["WISE", "2MASS"],      # Exclude specific catalogs
    exclude_freq_ranges=["radio"],           # Exclude frequency ranges
    x_axis="freq_ev",                        # Convert frequency to eV
    y_axis="flux_ev",                        # Convert flux to eV/cm²/s
)

Available axis conversions:

x_axis	Description
"freq_ev"	Hz to eV

y_axis	Description
"flux_ev"	erg/cm²/s to eV/cm²/s
"flux_norm"	Normalized eV/cm²/s
"flux_jyhz"	Jy·Hz
"flux_wm2"	W/m²
"nufnu_fnu_jy"	Fν in Jy

Source Metadata

info = client.sed.get_info(job.uuid)
print(info.source_name)   # "3C 273"
print(info.ra, info.dec)  # Coordinates
print(info.redshift)      # Redshift (float or None)
print(info.gal_lat)       # Galactic latitude
print(info.gal_long)      # Galactic longitude
print(info.W_peak)        # Peak frequency indicator

Download as CSV

client.sed.download_csv(job.uuid, "sed_data.csv")

# With filters (same options as get_data)
client.sed.download_csv(
    job.uuid, "filtered.csv",
    mjd_start=55000.0,
    exclude_catalogs=["WISE"],
)

CSV columns: frequency, flux, flux_err, MJD_start, MJD_end, flag, catalog, reference

---

Observations

Direct access to the unified observations catalog — 12.4M rows of multi-wavelength data spanning both per-frequency SED measurements and time-series lightcurves, distinguished by an is_lightcurve flag.

Catalogs: MMDCGR (Fermi γ-ray), MMDCOUV (Swift UVOT), MMDCXRT (Swift XRT), MMDCNuX (NuSTAR), ASAS-SN, ZTF, PanSTARRS-LC, SMARTS.

Filtered Query

# SED rows only, X-ray catalog, latest 100 by mjd_mid
rows = client.observations.query(
    catalog="MMDCXRT",
    is_lightcurve=False,
    limit=100,
)
for r in rows:
    print(f"{r.catalog} obsid={r.obsid} freq={r.frequency:.2e} flux={r.flux:.2e}")

# ZTF R-band lightcurve within an MJD window
lc = client.observations.query(
    catalog="ZTF",
    is_lightcurve=True,
    filter_band="R",
    mjd_min=60000,
    mjd_max=60100,
    ordering="-mjd_mid",
)

Cone Search

HEALPix-indexed spatial search (5 arcsec default, server caps prefilter at 4096 pixels):

# All observations within 10″ of 3C 273
near = client.observations.cone_search(
    ra=187.27791667,
    dec=2.05238889,
    radius_arcsec=10,
)

# Combine cone search with other filters
recent_lc = client.observations.cone_search(
    ra=187.27791667, dec=2.05238889, radius_arcsec=30,
    is_lightcurve=True,
    mjd_min=60000,
)

Available Filters (on both query() and cone_search())

Param	Type	Description
catalog	str	Catalog code (see list above)
filter_band	str	Photometric band (R, V, G, W1, …)
is_lightcurve	bool	True=LC only, False=SED only, omit=both
is_upper_limit	bool	Filter by upper-limit flag
obsid	str	Exact telescope observation ID
mjd_min, mjd_max	float	MJD range bounds
ordering	str	Sort key, - prefix for descending
limit	int	Cap returned rows

Reading Results

Each Observation carries its own is_lightcurve flag so a mixed result can be split client-side:

rows = client.observations.query(catalog="MMDCGR", limit=200)
lc  = [r for r in rows if r.is_lightcurve]
sed = [r for r in rows if not r.is_lightcurve]

SED rows carry frequency/mjd_start/mjd_end, LC rows carry mjd_mid/filter_band. The MMDC* catalogs auto-attach the Sahakyan et al. 2024 reference:

row = client.observations.query(catalog="MMDCXRT", limit=1)[0]
if row.reference:
    print(row.reference.bibcode)  # "2024AJ....168..289S"
    print(row.reference.citation) # "Sahakyan N., et al., 2024, ..."

---

Swift UVOT/XRT Analysis

client.madam runs the MADAM pipeline on Swift data for a sky position, on demand. Results are ingested into the observations catalog (MMDCOUV for UVOT, MMDCXRT / MMDCXRT_ORBIT for XRT), so anything already analysed at that position is answered from cache without a new run.

Ask for a time window (MJD) or a list of Swift obsids, never both:

# Time-range mode
job = client.madam.analyze(
    ra=166.1138, dec=38.2088,
    mjd_start=58849.0, mjd_end=59031.0,
    source_name="Mkn 421",
)

# ObsID mode (also the way to reach pre-2008 Swift data)
job = client.madam.analyze(ra=133.703645, dec=20.108511, obsids=["00030901030"])

print(job.status)   # "done" or "no_data" — both are complete answers
print(job.count)    # rows matching the request

job.results is a list of Observation objects, the same shape as client.observations returns. Each measurement appears twice: as a lightcurve row (is_lightcurve=True, mjd_mid, filter_band) and as an SED row (is_lightcurve=False, frequency, mjd_start/mjd_end).

sed = [r for r in job.results if not r.is_lightcurve]
lc  = [r for r in job.results if r.is_lightcurve]

Instruments

instrument	Runs	Catalogs
"uvot" (default)	UVOT photometry in U, B, V, W1, M2, W2	MMDCOUV
"xrt"	XRT spectroscopy, per observation and optionally per orbit	MMDCXRT, MMDCXRT_ORBIT
"swift"	both	all three

job = client.madam.analyze(
    ra=166.1138, dec=38.2088, obsids=["00030352001", "00030352002"],
    instrument="xrt",
    orbit=False,     # skip the per-orbit products; False is only accepted with instrument="xrt"
)
print(job.xrt_fits)  # spectral fit table, one dict per obsid; not part of job.results

X-ray runs are capped server-side at 60 obsids or a 365-day window per request, because XRT costs minutes per observation.

Coverage cache

The server remembers every completed request. A new request that lies inside an earlier one at the same position (5″) and with a covering instrument is answered immediately with cached=True, including the case where the earlier run found nothing. A "swift" run covers later "uvot" and "xrt" requests; the reverse is not true. Pass force=True to re-run regardless.

On a cache hit analyze() returns a job with cached=True whose request fields (instrument, mjd_start/mjd_end, obsids) and results are yours, while the run metadata (created_at, rows_ingested, xrt_fits, ...) comes from the covering run. Pass limit= to cap the rows either way.

Windows that reach up to today are handled sensibly: the last ~7 days are treated as not yet in the Swift archive, so a cached window still counts as covering them.

Timing

A run downloads Swift data and runs HEASoft, so it takes minutes to hours. analyze() polls once a minute, backing off to every two minutes, and gives up after four hours by default:

job = client.madam.analyze(
    ra=166.1138, dec=38.2088, mjd_start=58849.0, mjd_end=59031.0,
    poll_interval=60.0,   # seconds between checks (keep >= 60)
    max_minutes=240.0,
)

If the pipeline fails, analyze() and wait_for_completion() raise AnalysisJobError carrying the tail of the container log in .logs. If the wait budget runs out, PollingTimeoutError.uuid identifies the job, which keeps running on the server and can be collected later with client.madam.get(uuid).

---

Blazar Emission Modeling

MMDC supports three blazar broadband emission models:

Model	Description
SSC	Synchrotron Self-Compton
EIC	External Inverse Compton
HADRONIC	Hadronic emission model

Input Data Format

All modeling endpoints expect a CSV file with three columns (case-sensitive, lowercase):

frequency,flux,flux_err
1.00e+09,2.50e-14,3.00e-15
4.85e+09,3.10e-14,2.80e-15
...

Validate CSV Before Submitting

validation = client.modeling.validate_csv("observations.csv")

print(validation.success)          # True/False
print(validation.data_points)      # Number of valid rows
print(validation.columns)          # ["frequency", "flux", "flux_err"]
print(validation.frequency_range)  # [min, max]
print(validation.flux_range)       # [min, max]

SSC Model Fitting

result = client.modeling.batch_infer(
    "observations.csv",
    z=0.158,            # Redshift (0 < z <= 10)
    ebl=True,           # EBL absorption correction
    model_type="SSC",
)

print(result.pdf_link)                   # URL to PDF report
print(result.csv_best_parameters_link)   # URL to best-fit parameters CSV
print(result.csv_best_model_link)        # URL to best-fit model CSV
print(result.best_parameters)            # Dict of parameter name -> {value, error}

Fixed Parameters

Fix specific model parameters instead of fitting them:

result = client.modeling.batch_infer(
    "observations.csv",
    z=0.158,
    ebl=True,
    model_type="SSC",
    fixed_parameters={
        "log_B": -1.5,
        "lorentz_factor": 20.0,
    },
)

SSC parameters: log_B, log_electron_luminosity, log_gamma_cut, log_gamma_min, log_radius, lorentz_factor, spectral_index

EIC parameters: log_B, log_Ld, log_MBH, log_electron_luminosity, log_gamma_cut, log_gamma_min, log_radius, lorentz_factor, spectral_index, log_nu_BLR, log_nu_DT

HADRONIC parameters: log_B, log_Le, log_gamma_e_cut, log_gamma_e_min, log_gamma_p_cut, log_Lp, log_R, lorentz_factor, pe, pp

EIC Model Fitting

result = client.modeling.batch_infer(
    "observations.csv",
    z=0.5,
    ebl=True,
    model_type="EIC",
    fixed_parameters={"log_nu_BLR": 15.0, "log_nu_DT": 13.5},
)

HADRONIC Model with Neutrino Parameters

Hadronic models require additional neutrino likelihood parameters. Choose either Poisson or chi-square likelihood:

Poisson likelihood:

result = client.modeling.batch_infer(
    "observations.csv",
    z=1.0,
    ebl=True,
    model_type="HADRONIC",
    likelihood_type="poisson",
    n_icecube=3,       # Number of IceCube neutrino events
    dt=12.0,           # Observation period in months
)

Chi-square likelihood:

result = client.modeling.batch_infer(
    "observations.csv",
    z=1.0,
    ebl=True,
    model_type="HADRONIC",
    likelihood_type="chi2",
    x1=100.0,          # First neutrino energy (TeV)
    x2=200.0,          # Second neutrino energy (TeV)
    y=-12.0,           # Neutrino flux log value
)

Working with Results

result = client.modeling.batch_infer(...)

# Best-fit parameters
for name, param in result.best_parameters.items():
    print(f"{name}: {param.value} +/- {param.error}")

# Fixed parameters
if result.fixed_parameters:
    for name, param in result.fixed_parameters.items():
        print(f"{name} (fixed): {param.value}")

# Download links
print(result.pdf_link)                   # PDF report with plots
print(result.csv_best_parameters_link)   # Best-fit parameters as CSV
print(result.csv_best_model_link)        # Best-fit model curve as CSV

Download result files:

import httpx

if result.pdf_link:
    pdf = httpx.get(result.pdf_link)
    with open("report.pdf", "wb") as f:
        f.write(pdf.content)

Error Handling

from astro_mmdc import (
    MMDC,
    MMDCError,           # Base exception for all SDK errors
    APIError,            # Non-2xx HTTP response
    NotFoundError,       # 404 response
    ValidationError,     # 422 response (CSV/parameter validation)
    PollingTimeoutError, # Polling exceeded max wait time
    BatchJobError,       # Batch modeling job failed on the server
    AnalysisJobError,    # Swift/MADAM analysis job failed on the server
)

client = MMDC()

try:
    result = client.modeling.batch_infer("data.csv", z=0.5, ebl=True, model_type="SSC")
except ValidationError as e:
    print(f"CSV validation failed: {e} (type: {e.validation_type})")
except PollingTimeoutError:
    print("Job did not complete in time")
except NotFoundError:
    print("Resource not found")
except APIError as e:
    print(f"HTTP {e.status_code}: {e.detail}")

The SDK automatically retries on transient errors (429, 502, 503, 504) with exponential backoff (up to 3 attempts).

---

Advanced

Manual Job Control

SED preparation, batch modeling and Swift analysis are asynchronous — you submit a job, then poll for results. The convenience methods (prepare_and_wait, batch_infer, analyze) handle polling automatically, but you can manage each step yourself for more control.

This is useful when you want to submit multiple jobs at once and poll them independently, or do other work between submission and result retrieval.

SED: Manual Submit and Poll

# Submit — returns immediately
job = client.sed.prepare(ra=187.28, dec=2.05, database_name="3C273")
print(job.uuid)

# Check status manually
status = client.sed.get_status(job.uuid)
print(status.status)  # "processing", "done", "no_data", or "error"

# Or block until complete with custom polling settings
completed = client.sed.wait_for_completion(
    job.uuid,
    poll_interval=5.0,   # Seconds between checks
    max_minutes=15.0,    # Give up after this
)

Modeling: Manual Submit and Poll

# Submit — returns immediately with a batch_result_id
submission = client.modeling.submit_batch(
    "observations.csv",
    z=0.158,
    ebl=True,
    model_type="SSC",
)
print(submission.batch_result_id)

# Check result manually
result = client.modeling.get_batch_result(submission.batch_result_id)
if result.pdf_link:
    print("Job complete!")
else:
    print("Still processing...")

# Or block until complete with custom polling settings
result = client.modeling.wait_for_batch(
    submission.batch_result_id,
    poll_interval=10.0,
    max_minutes=15.0,
)

Swift Analysis: Manual Submit and Poll

# Submit — returns immediately
submission = client.madam.submit(ra=166.1138, dec=38.2088, mjd_start=58849.0, mjd_end=59031.0)
if submission.cached:
    rows = submission.results        # already analysed, no run queued
else:
    print(submission.uuid, submission.status)   # "processing"

# Poll cheaply without the rows
status = client.madam.get(submission.uuid, include_results=False)
print(status.status)  # "processing", "done", "no_data", or "error"

# Or block until complete
job = client.madam.wait_for_completion(submission.uuid, poll_interval=60.0, max_minutes=240.0)

# Fetch rows later, optionally capped
job = client.madam.get(submission.uuid, limit=100)

Submitting the same request twice while it is still running returns the running job rather than queuing a second one.

Batch Processing Multiple Sources

import time

sources = [
    {"ra": 187.28, "dec": 2.05, "name": "3C273"},
    {"ra": 166.11, "dec": 38.21, "name": "Mkn421"},
    {"ra": 253.47, "dec": 39.76, "name": "Mkn501"},
]

# Submit all jobs first
jobs = []
for src in sources:
    job = client.sed.prepare(ra=src["ra"], dec=src["dec"], database_name=src["name"])
    jobs.append(job)
    print(f"Submitted {src['name']}: {job.uuid}")

# Then wait for all of them
for job in jobs:
    completed = client.sed.wait_for_completion(job.uuid)
    print(f"{completed.source_name}: {completed.status}")

Synchronous Inference

For quick model calculations without queuing — pass parameters directly and get the spectrum back instantly:

result = client.modeling.infer(
    z=0.158,
    ebl=True,
    model_type="SSC",
    parameters={
        "log_B": -1.5,
        "log_electron_luminosity": 44.0,
        "log_gamma_cut": 5.0,
        "log_gamma_min": 2.0,
        "log_radius": 16.0,
        "lorentz_factor": 20.0,
        "spectral_index": 2.2,
    },
)

# Access the model spectrum directly
print(result.nu)       # Frequency values
print(result.nuFnu)    # Flux values (nu * F_nu)

Works with all model types — SSC, EIC, and hadronic:

# EIC inference
result = client.modeling.infer(
    z=0.5,
    ebl=True,
    model_type="EIC",
    parameters={
        "log_B": -1.0,
        "log_electron_luminosity": 44.0,
        "log_gamma_cut": 4.5,
        "log_gamma_min": 2.0,
        "log_radius": 16.5,
        "lorentz_factor": 15.0,
        "spectral_index": 2.0,
        "log_Ld": 45.0,
        "log_MBH": 8.5,
        "log_nu_BLR": 15.0,
        "log_nu_DT": 13.5,
    },
)

BatchResult Object Reference

All fields available on a BatchResult:

result.data                       # Model curve data points (dict)
result.best_parameters            # Best-fit parameters (dict of name -> {value, error})
result.fixed_parameters           # Fixed parameters (dict of name -> {value, error})
result.model_type                 # "SSC", "EIC", or "HADRONIC"
result.z                          # Redshift
result.pdf_link                   # URL to PDF report
result.csv_best_parameters_link   # URL to best-fit parameters CSV
result.csv_best_model_link        # URL to best-fit model curve CSV
result.uploaded_file              # Original input data (dict)
result.multinest_stats            # MultiNest sampling statistics (dict)
result.equal_weighted_posterior   # Posterior samples (dict)

---

API Reference

client.sed

Method	Description
prepare(ra, dec, database_name, ...)	Submit SED preparation job
get_status(uuid)	Check job status
wait_for_completion(uuid, ...)	Poll until job finishes
prepare_and_wait(ra, dec, database_name, ...)	Submit and wait
get_data(uuid, ...)	Get frequency/flux data
get_info(uuid)	Get source metadata
download_csv(uuid, dest, ...)	Download data as CSV file

client.modeling

Method	Description
validate_csv(file)	Validate CSV before submission
submit_batch(file, z, ebl, model_type, ...)	Submit batch inference job
get_batch_result(batch_result_id)	Get current job result
wait_for_batch(batch_result_id, ...)	Poll until job completes
batch_infer(file, z, ebl, model_type, ...)	Submit and wait
infer(z, ebl, model_type, parameters)	Synchronous model inference
csv_to_json(file)	Convert CSV to JSON format

client.observations

Method	Description
query(catalog, filter_band, is_lightcurve, ...)	Filtered query of the observations table
cone_search(ra, dec, radius_arcsec, ...)	HEALPix-indexed spatial search with optional filters

client.madam

Method	Description
submit(ra, dec, mjd_start=, mjd_end=, obsids=, instrument=, ...)	Request a Swift analysis; cached answers come back with rows
get(uuid, include_results=True, limit=None)	Fetch a job, optionally without its rows
wait_for_completion(uuid, poll_interval=60, max_minutes=240, limit=None)	Poll until done / no_data; raises AnalysisJobError on error
analyze(ra, dec, ..., limit=None, poll_interval=60, max_minutes=240)	Submit and wait

---

Links

• MMDC Platform: mmdc.am
• Source Code: github.com/ICRANet/mmdc
• Issues: github.com/ICRANet/mmdc/issues