QuTiP Development Roadmap¶
This document outlines plan and ideas for the current and future development of QuTiP. The document is maintained by the QuTiP Admim team. Contributuions from the QuTiP Community are very welcome.
In particular this document outlines plans for the next major release of qutip, which will be version 5. And also plans and dreams beyond the next major version.
There is lots of development going on in QuTiP that is not recorded in here. This a just an attempt at coordinated stragetgy and ideas for the future.
What is QuTiP?¶
The name QuTiP refers to a few things. Most famously, qutip is a Python library for simulating quantum dynamics. To support this, the library also contains various software tools (functions and classes) that have more generic applications, such as linear algebra components and visualisation utilities, and also tools that are specifically quantum related, but have applications beyond just solving dynamics (for instance partial trace computation).
QuTiP is also an organisation, in the Github sense, and in the sense of a group of people working collaboratively towards common objectives, and also a web presence qutip.org. The QuTiP Community includes all the people who have supported the project since in conception in 2010, including manager, funders, developers, maintainers and users.
These related, and overlapping, uses of the QuTiP name are of little consequence until one starts to consider how to organise all the software packages that are somehow related to QuTiP, and specifically those that are maintained by the QuTiP Admim Team. Herin QuTiP will refer to the project / organisation and qutip to the library for simulating quantum dyanmics.
Should we be starting again from scratch, then we would probably chose another name for the main qutip library, such as qutip-quantdyn. However, qutip is famous, and the name will stay.
Library package structure¶
With a name as general as Quantum Toolkit in Python, the scope for new code modules to be added to qutip is very wide. The library was becoming increasingly difficult to maintain, and in c. 2020 the QuTiP Admim Team decided to limit the scope of the ‘main’ (for want of a better name) qutip package. This scope is restricted to components for the simulation (solving) of the dynamics of quantum systems. The scope includes utilities to support this, including analysis and visualisation of output.
At the same time, again with the intention of easing maintence, a decision to limit dependences was agreed upon. Main qutip runtime code components should depend only upon Numpy and Scipy. Installation (from source) requires Cython, and some optional components also require Cython at runtime. Unit testing requires Pytest. Visualisation (optional) components require Matplotlib.
Due to the all encompassing nature of the plan to abstract the linear algebra data layer, this enhancement (developed as part of a GSoC project) was allowed the freedom (potential for non-backward compatibility) of requiring a major release. The timing of such allows for a restructuring of the qutip compoments, such that some that could be deemed out of scope could be packaged in a different way – that is, not installed as part of the main qutip package. Hence the proposal for different types of package described next. With reference to the discussion above on the name QuTiP/qutip, the planned restructuring suffers from confusing naming, which seems unavoidable without remaining either the organisation or the main package (neither of which are desirable).
- QuTiP family packages
The main qutip package already has sub-packages, which are maintained in the main qutip repo. Any packages maitained by the QuTiP organisation will be called QuTiP ‘family’ packages. Sub-packages within qutip main will be called ‘integrated’ sub-packages. Some packages will be maintained in their own repos and installed separately within the main qutip folder structure to provide backwards compatibility, these are (will be) called qutip optional sub-packages. Others will be installed in their own folders, but (most likely) have qutip as a dependency – these will just be called ‘family’ packages.
- QuTiP affilliated packages
Other packages have been developed by others outside of the QuTiP organisation that work with, and are complementary to, qutip. The plan is to give some recognition to those that we deem worthy of such [this needs clarification]. These packages will not be maintained by the QuTiP Team.
Do we really need optional subpackages? It seems that are a bit fiddly and as we are side-stepping bw compat with a major version, we could just make QIP a separate package.
current package status: family package qutip
planned package status: family package qutip
The in-scope components of the main qutip package all currently reside in the base folder. The plan is to move some components into integrated subpackages as follows:
core quantum objects and operations
solver quantum dynamics solvers
What will remain in the base folder will be miscellaneous modules. There may be some opportunity for grouping some into a visualisation subpackage. There is also some potential for renaming, as some module names have underscores, which is unconventional.
current package status: integrated sub-package qutip.control
planned package status: family package qtrl
There are many OSS Python packages for quantum control optimisation. There are also many different algorithms. The current control integrated subpackage provides the GRAPE and CRAB algorithms. It is too ambitious for QuTiP to attempt (or want) to provide for all options. Control optimisation has been deemed out of scope and hence these components will be separated out into a family package called Qtrl.
Potentially Qtrl may be replaced by separate packages for GRAPE and CRAB, based on the QuTiP Control Framework.
current package status: integrated sub-package qutip.qip
planned package status: optional sub-package qutip.qip
Is it really necessary for this to be a sub-package? It could just be a separate package.
The QIP subpackage has been deemed out of scope (feature-wise). It also depends on qutip.control and hence would be out of scope for dependency reasons. A separate repository has already been made for qutip-qip.
current package status: independent package sympsi
planned package status: family package qutip-symbolic
Long ago Robert Johansson and Eunjong Kim developed Sympsi. It is a fairly coomplete library for quantum computer algebra (symbolic computation). It is primarily a quantum wrapper for Sympy.
It has fallen into unmaintained status. The latest version on the sympsi repo does not work with recent versions of Sympy. Alex Pitchford has a fork that does ‘work’ with recent Sympy versions – unit tests pass, and most examples work. However, some (important) examples fail, due to lack of respect for non-commuting operators in Sympy simplifcation functions (note this was true as of Nov 2019, may be fixed now).
There is a [not discussed with RJ & EK] plan to move this into the QuTiP family to allow the Admin Team to maintain, develop and promote it. The ‘Sympsi’ name is cute, but a little abstract, and qutip-symbolic is proposed as an alternative, as it is plainer and more distinct from Sympy.
data layer abstraction¶
Development completed as a GSoC project. Fully implemented in the dev.major branch. Currently being used by some research groups.
Abstraction of the linear algebra data from code qutip components, allowing for alternatives, such as sparse, dense etc. Difficult to summarize. Almost every file in qutip affected in some way. A major milestone for qutip. Significant performance improvements throughout qutip.
Some developments tasks remain, including providing full control over how the data-layer dispatchers choose the most appropriate output type.
qutip main reorganization¶
Reorganise qutip main components to the structure described above.
qutip user docs migration¶
- admin lead
- main dev
The qutip user documentation build files are to be moved to the qutip/qutip repo. This is more typical for an OSS package.
As part of the move, the plan is to reconstruct the Sphinx structure from scratch. Historically, there have been many issues with building the docs. Sphinx has come a long way since qutip docs first developed. The main source (rst) files will remain [pretty much] as they are, although there is a lot of scope to improve them.
The qutip-doc repo will afterwards just be used for documents, such as this one, pertaining to the QuTiP project.
Solver data layer integration¶
The new data layer gives opportunity for significantly improving performance of the qutip solvers. Eric has been revamping the solvers by deploying QobjEvo (the time-dependent quantum object) that he developed. QobjEvo will exploit the data layer, and the solvers in turn exploit QobjEvo.
A separate package for qutip-qip was created during Sidhant’s GSoC project. There is some fine tuning required, especially after qutip.control is migrated.
An overhaul of the HEOM solver. C++ components used to speed up construction of the hierarchy.
- admin lead
- main dev
The components currently packaged as an integrated subpackage of qutip main will be moved to separate package called Qtrl. This is the original codename of the package before it was integrated into qutip. Also changes to exploit the new data layer will be implemented.
QuTiP control framework¶
- admin lead
- main dev
Create new package qutip-ctrlfw “QuTiP Control Framework”. The aim is provide a common framework that can be adopted by control optimisation packages, such that different packages (algorithms) can be applied to the same problem.
Classes for defining a controlled system:
named control parameters. Scalar and n-dim. Continuous and discrete variables
mapping of control parameters to dynamics generator args
masking for control parameters to be optimised
Classes for time-dependent variable parameterisation
Classes for defining an optimisation problem:
single and multiple objectives
- admin lead
- main dev
A wrapper for multi-variable optimisation functions. For instance those in scipy.optimize (Nelder-Mead, BFGS), but also others, such as Bayesian optimisation and other machine learning based approaches. Initially just providing a common interface for quantum control optimisation, but applicable more generally.
- admin lead
- main dev
Create a new family package qutip-symbolic from ajgpitch fork of Sympy. Must gain permission from Robert Johansson and Eunjong Kim. Extended Sympy simplify to respect non-commuting operators. Produce user documentation.
Status messaging and recording¶
- admin lead
- main dev
QuTiP has various ways of recording and reporting status and progress.
ProgressBar used by some solvers
Python logging used in qutip.control
Dump used in qutip.control
heom records solver.Stats
Some consolidation of these would be good.
Some processes (some solvers, correlation, control optimisation) have many stages and many layers. Dump was initially developed to help with debugging, but it is also useful for recording data for analysis. qutip.logging_utils has been criticised for the way it uses Python logging. The output goes to stderr and hence the output looks like errors in Jupyter notebooks.
Clearly, storing process stage data is costly in terms of memory and cpu time, so any implementation must be able to be optionally switched on/off, and avoided completely in low-level processes (cythonized components).
optional recording (storing) of process stage data (states, operators etc)
optionally write subsets to stdout
maybe other graphical representations
option to save subsets to file
should ideally replace use of ProgressBar, Python logging, control.Dump, solver.Stats
- admin lead
- main dev
QuTiP is pretty simple to use at an entry level for anyone with basic Python skills. However, some Python skills are necessary. A graphical user interface (GUI) for some parts of qutip could help make qutip more accessible. This could be particularly helpful in education, for teachers and learners.
This would make an good GSoC project. It is independent and the scope is flexible.
The scope for this is broad and flexible. Ideas including, but not limited to:
Interactive Bloch sphere¶
Matplotlib has some interactive features (sliders, radio buttons, cmd buttons) that can be used to control parameters. They are a bit clunky to use, but they are there. Could maybe avoid these and develop our own GUI. An interactive Bloch sphere could have sliders for qubit state angles. Buttons to add states, toggle state evolution path.
Options to configure dynamics generators (Lindbladian / Hamiltonian args etc) and expectation operators. Then run solver and view state evolution.
QIP circuits could be animated. Status lights showing evolution of states during the processing. Animated Bloch spheres for qubits.
QuTiP major release roadmap¶
These Projects need to be completed for the qutip v.5 release.
The planned timeline for the release is:
alpha version, April 2021. Core features packaged and available for experienced users to test.
beta version, July 2021. All required features and documentation complete, packaged and ready for community testing.
full release, September 2021. Full tested version released.