# API References

TorchSim is organized around one idea: a **signal model** is the only thing a
sequence has to supply, and everything else -- differentiation, execution
across devices, estimation, reconstruction, design -- is written once against
that interface.

The pages below follow the subpackages.

{doc}`model`
: What a sequence of your own is written from: the physics, saying
  what each kind of event does to the spins, and a simulator saying what
  order they are played in.

{doc}`simulators`
: The sequences that ship with TorchSim, as classes and as one-call
  functional wrappers.

{doc}`sequence`
: The device-agnostic description an acquisition is assembled from: events,
  the operators that emit them, and builders for whole sequences.

{doc}`estimators`
: Estimating tissue properties from a measured volume: dictionary matching,
  lookup tables, nonlinear least squares, kernel regression.

{doc}`recon`
: Solving for parameter maps straight from k-space, with the signal model
  inside the forward operator, and the temporal basis that makes it cheap.

{doc}`optim`
: Choosing a sequence's parameters by minimizing a cost you write.

{doc}`execution`
: Running a description: the differentiable engine, transmit calibration,
  and where a per-voxel workload is placed.

```{toctree}
:hidden:
:maxdepth: 1

model
simulators
sequence
estimators
recon
optim
execution
```