Skip to main content

Frequently Asked Questions

What is CSDL?#

CSDL is an embedded domain specific language designed for solving Multidisciplinary Design Analysis and Optimization (MDAO) problems. CSDL enables complete separation between specifying a mathematical model and implementing a numerical simulation of a physical system. Separation between specification and implementation enables engineers to operate at a high level of abstraction, without the need to implement low level algorithms, including derivative computation. CSDL is an obvious choice for optimization problems that benefit from gradient-based methods because of how CSDL automates derivative computation, both at the level of individual operations, and across tightly coupled disciplines. The CSDL compiler's intrusive design makes it a powerful tool for analyzing, verifying, and optimizing model code prior to execution, so it is sure to grow in capability in the long term, reducing development time.

How do I install csdl?#

Please follow the instructions in the Tutorial.

How is CSDL licensed?#

CSDL is licensed under the Lesser GNU Public License, version 3. The license is included in the source code for CSDL. A summary of the license is also available here.

Why use CSDL?#

CSDL is a powerful language that enables users to specify a mathematical model of a physical system without regard for how to implement a simulation for that specification, or which architecture is used to run the simulation. CSDL is especially powerful in multidisciplinary design optimization where gradient-based approaches can be applied, due to their superior performance over gradient-free approaches. Many modeling languages exist that excel at solving general nonlinear programs, but those languages are typically restricted to a single discipline, do not fully automate derivative computation, or cannot compute the derivatives required for optimization efficiently. This leads to overhead for the user to specify a model as a single discipline, or implement derivative computation which may be difficult to maintain. CSDL automates derivative computation for individual operations and across multiple disciplines. Model disciplines are typically tightly coupled, presenting a challenge for derivative computation necessary for solving nonlinear programs (NLPs).

CSDL enables users to specify mathematical models of physical systems in two ways: a functional style, and an object oriented style. The functional style of model specification is a style that is similar to natural mathematical notation. The object oriented style of model specification enables hierarchical composition of models. Both styles make code easier to read and intuitive to use for multidisciplinary design analysis and optimization applications.

What is the goal of this project?#

The goal of CSDL is to separate model specification from simulation implementation, eliminating overhead imposed on the model developer, with emphasis given to solving multidisciplinary design optimization problems. For example, CSDL automatically computes derivatives (necessary for gradient-based optimization) across multiple disciplines without any additional code from the user. Complete separation between model specification and simulation implementation allows for the language to evolve to fit user needs with little or no additional overhead for the user.

Who is CSDL for?#

CSDL is targeted at Multidisciplinary Design Optimization (MDO) researchers and practitioners, and as a result, is general enough for anyone who wishes to create a mathematical model of a physical system and simulate the behavior of that system, even if gradient-based optimization is not a requirement. CSDL is especially powerful in MDO, where the model is composed of multiple disciplines that are tightly coupled, which presents a challenge to efficient derivative computation. The tight coupling between disciplines requires computing derivatives of the objective and constraints with respect to design variables in order for an optimizer to update the values of the design variables. CSDL comes with an extensive standard library that automates derivative computation. If the Standard Library is insufficient for the needs of a project, users can also define their own operations and the corresponding partial derivative for each operation as well using custom operations.

Who uses CSDL?#

Currently, the LSDO Lab at UCSD uses CSDL for a number of projects. If you are a CSDL user and would like your organization or project to be featured on this website, please let us know on Zulip.

Who maintains CSDL?#

The LSDO Lab at UCSD owns the repository for the csdl compiler front end source code and the csdl_om compiler back end source code. At the moment, Victor Gandarillas is the sole maintainer for both csdl and the csdl_om compiler back end (which is based on OpenMDAO), but more contributors are welcome. As CSDL grows and evolves, we may set up various working groups to ensure that CSDL meets the needs of our users.

Do I need to use the command line to compile my program?#

Users write CSDL source code, compile source code, and run the generated executable from within a single Python script Running the Python script containing the CSDL source code and commands to compile and run a simulation is the only step required. Although CSDL is a compiled language, the compilation (creating a Simulator object) and execution (calling Simulator.run) steps are separated by lines of code within a script, rather than two separate commands on a terminal/command line (e.g. for a simple program written in C, gcc main.c followed by ./a.out).

Why do I need to install two packages to install the CSDL compiler?#

CSDL is a language, and the CSDL compiler is implemented as a three stage compiler, which consists of a front end that generates an intermediate representation, a middle end that performs implementation independent optimizations on the intermediate representation, and a back end, which generates code in a target language, or generates an executable directly. CSDL is an embedded domain specific language (EDSL), which means that CSDL is basically a language within a language; Python is the host language for CSDL. The csdl package provides the front end and middle end of the CSDL compiler, but the user is free to choose any CSDL compiler back end to implement a low level executable object to run a simulation. The flexibility of choosing any back end means that all code written in CSDL could in theory run on any hardware, and implementation specific code optimizations can be performed by the CSDL compiler without any change to user code, the csdl package, or the design of the CSDL language. Choosing an implementation for a project is as simple as changing one line of code, from <backend> import Simulator, where <backend> (including angle brackets) is replaced with the name of the package that implements the Simulator class.

What is the difference between CSDL, csdl, and the "back end"?#

CSDL is a language for specifying a model of a physical system whose behavior a user wants to simulate for the purpose of analysis or optimization. Like any language, CSDL requires an implementation (compiler or interpreter) in order to generate and run software on a computer. The csdl package contains the front end and middle end of the CSDL compiler. The csdl package does not provide the complete implementation of the CSDL compiler; a separate package is required to supply the compiler back end. All back ends must provide a Simulator class, which provides the same API as the SimulatorBase class provided in the csdl package. Conformance to a single API enables users to switch between back ends by changing one line of code, from <backend> import Simulator, where <backend> (including angle brackets) is replaced with the name of the package that implements the Simulator class.

Where can I run code written using CSDL?#

You may run CSDL on any platform that your chosen CSDL compiler back end supports. Consult documentation for the compiler back end of your choice for more information. A list of available compiler back ends is available here.

Where is CSDL headed?#

CSDL is a very young project, so the short answer is, that's up to you! We are currently using CSDL in a variety of projects, gathering user feedback, and working with potential stakeholders to make sure CSDL continues to grow, benefit the MDO community, and make lasting impact. Feel free to use CSDL in your project, send us feedback, and engage with the community.

How can I get involved?#

Please see the Community page for more information on how to get in touch with other users and developers. Please see the section on contributing to CSDL in the Developer Docs.