Highlights of COPT 7.0, the Sequoia Number Solver
① The mixed integer programming solver has achieved a breakthrough improvement, with its performance comprehensively surpassing CPLEX. Cordless Vacuum Cleaner China
② In practical applications across multiple fields, the MIP solver has made significant progress, with a solution speed increase of ten to one hundred times compared to last year's version 5.0.
③ The performance of the second-order cone programming solver has been greatly improved, surpassing Mosek and ranking first in the world in the ASU evaluation.
④ A new online licensing (Web License) authorization method has been added, supporting cloud deployment.
The latest version of the Sequoia Solver, COPT 7.0 , has been officially released. This upgraded version significantly improves the performance of the mixed integer programming and second-order cone programming solvers, further enhancing the speed and stability of linear programming and other solution modules. It has achieved six first-place rankings and four second- rankingsplace on the ASU benchmark platform, a third-party mathematical optimization software evaluation platform. Specifically, it ranks first in the world for both the optimal numerical solution and optimal vertex solution rankings for linear programming, first in the world for the SOCP ranking, and second in the world for the MIP ranking.
At the same time, COPT 7.0 adds an online licensing (Web License) method, which broadens the deployment options of COPT and provides a flexible and quick choice for cloud-based usage scenarios such as enterprise production.
New users are welcome to apply for a trial of the latest version of COPT on the Sugon website . Existing users need to reinstall COPT and update the license file. As a way of giving back to our users, academic users will receive another 365-day free trial upon re-application.
A breakthrough has been achieved in the performance of the Mixed Integer Programming ( MIP ) solver , which now comprehensively outperforms CPLEX.
Mixed-integer programming (MIP) solvers have wide applications and are far more difficult to develop than other modules such as linear programming. Therefore, MIP performance is a crucial standard for evaluating optimization software performance . Since its release, COPT has been dedicated to the research and development of this module, finally achieving a breakthrough . According to recent evaluation results released by "Data Magician," the latest version 7.0 of COPT significantly outperforms the latest version 22.1 of CPLEX from European and American manufacturers. It can solve a larger number of problems and its solution time is 33% faster, marking a milestone in the development of domestically produced solvers.
The following chart shows the performance evaluation results of various iterations of COPT on the MIPLIB benchmark dataset. The number of unsolved problems is gradually decreasing, and the speedup ratio is gradually increasing. The Sugon solver team will continue to strive for further progress, and we welcome everyone to stay tuned!
Note:
1. Data source: Benchmarks for Optimization Software by Prof. Hans Mittlemann. (https://plato.asu.edu/bench.html)
2. "Number of unsolved problems" indicates the number of problems that COPT failed to solve in each update of the MIPLIB 2017 test set (240 test cases);
3. "Speedup ratio" indicates the relative solution speed of each COPT update compared to version V2.0.
The continuous improvement of the Sequoia Solver's functionality and performance is attributed to its practical application in customer projects, accumulated experience, and ongoing parameter optimization. Beyond public benchmark rankings, the latest version of COPT, 7.0, has also demonstrated significant performance improvements in real-world applications across various fields, achieving a tenfold to a hundredfold increase compared to last year's COPT 5.0. Below is a comparison of the solution times for different COPT versions in real-world MIP examples:
The SOCP solver has seen a significant performance improvement, ranking first in the world in the ASU benchmark.
The ASU benchmark rankings, maintained by Professor Hans Mittelmann of Arizona State University, are a third-party, internationally authoritative benchmarking platform for mathematical optimization software. The benchmarks use representative as examples, and their evaluations are widely recognized in the solver industry. COPT 7.0 ranks highly on the ASU benchmark rankings, achieving 6 world firsts and 4 world seconds. Notably, for the second-order cone programming (SOCP) problem, which has important and widespread applications in the financial sector , COPT 7.0 has significantly improved its solving performance, surpassing the Danish company Mosek, which has specialized in this area for many years, to reach the world's top level . This is another breakthrough for COPT in nonlinear modules, following its MIQP module in version 6.5 surpassing Gurobi to achieve first place. The following will introduce the benchmark results for all COPT solving modules on this list:
1. The linear programming module maintains its leading position, ranking first in the world in both the optimal vertex solution and optimal numerical solution leaderboards.
COPT 7.0 ranks first in both the LP-optimal vertex solution leaderboard and the LP-optimal numerical solution leaderboard.
2. The MIP module has been significantly improved, further narrowing the gap with the first place.
COPT version 7.0 has achieved a breakthrough , ranking second in the world in all three MIP benchmark lists, and further narrowing the gap with the first place in overall solving performance.
COPT 7.0 Mixed Integer Programming (MIP) Module Evaluation Results
3. The nonlinear programming module now includes SOCP, a world first.
In the nonlinear programming solution module, COPT 7.0 achieved four world firsts and one world second in the evaluation results. Among them, the second-order cone programming module surpassed Mosek, winning first place in the test list.
Evaluation results of the COPT 7.0 second-order cone programming (SOCP) solver module.
The following table summarizes the types of problems that COPT 7.0 supports and the latest performance evaluation results :
Note: The data in the table above comes from the evaluation results on October 17, 2023 , from the internationally authoritative mathematical optimization software evaluation platform maintained by Professor Hans Mittelmann of Arizona State University, http://plato.asu.edu/bench.html.
A new online licensing method has been added, supporting cloud deployment.
COPT's Web License provides users with remote licensing services. Regardless of whether the client is located in the cloud or in a container, as long as it can access the Internet via HTTPS, it can obtain a token from COPT's Web License licensing server to run COPT. No hardware binding is required, and cross-version usage is supported. Compared to existing traditional licensing methods, online licensing is not limited by fixed hardware environments, providing fast and flexible support for service migration, switching, and scaling in enterprise and university cloud production environments.
Meanwhile, corresponding to traditional authorization methods, online licensing also includes three subcategories: online server license (Web License-Server), online floating license (Web License-Floating), and online cluster license (Web License-Cluster).
1. Web License ( Server )
For individual trial and commercial users: Supports deploying and running COPT servers in the cloud (without binding to any machine's hardware information), and multiple modeling and solving tasks can be performed simultaneously on the server.
2. Online Floating License ( Web License-Floating )
For business users: Supports the deployment of a floating token server in the cloud as a server (this server needs to be connected to the Internet and obtain remote authorization through online licensing) to authorize other machines (clients) on the local area network to run COPT.
3. Online Cluster License ( Web License-Cluster )
For business users: Supports setting up one or more computing cluster servers in the cloud , allowing modeling to be performed on local machines (clients) within a local area network and solved on remote cluster servers (servers) to efficiently utilize the powerful computing resources of the servers.
We provide a web license client application at https://copt.shanshu.ai/license . After registering and logging in, you can directly apply for the three types of online licenses mentioned above, obtain the authorization configuration file, and manage token usage, machine usage, etc.
Web License web interface
In addition, to provide users with a more user-friendly experience, we have also prepared an online resource page for COPT, which will gather user guides, frequently asked questions, sample code, download updates, etc., presented in a clear modular format, making it easy for users to access information and get the latest product updates at any time. It will be launched on the Sugon website soon, so stay tuned!
Introduction to COPT (Cephalograph Number Solver )
COPT 7.0 can be used to efficiently and stably solve linear programming, mixed-integer programming, (mixed-integer) second-order cone programming, semidefinite programming, (mixed-integer) convex quadratic programming, and (mixed-integer) convex quadratic constrained programming problems. In addition, COPT provides rich and user-friendly auxiliary functions, mainly including:
1. Analysis functions for infeasibility problems: Calculate the minimum conflict set (Irreducible Inconsistent Subsystem, or IIS) of the infeasibility model, and calculate feasibility relaxation (FeasRelax).
2. Advanced control functions for MIP problems: setting the initial MIP solution and callback functions;
3. An automatic tuning tool for optimization parameters: COPT Tuner;
4. The COPT Python interface supports matrix modeling and generalized constraints, and supports installing and updating coptpy via pip. The provided coptpy-stubs support type hints.
COPT supports all major operating systems, such as Windows, macOS, and Linux (including Loongson architecture, Apple's own chips, and arm64 chips). Users can use COPT from various programming languages and modeling tools, including data-based C language interfaces, object-oriented Python, C++, C#, and Java interfaces, as well as third-party modeling tool interfaces such as Julia, AIMMS, AMPL, GAMS, Pyomo, PuLP, and CVXPY; it also supports Matlab and Matlab-Yalmip interfaces.
In addition, COPT supports multiple deployment methods. Besides personal computers and servers, we also offer advanced deployment options such as online licensing, floating licenses, and compute clusters, facilitating use in school labs and corporate production environments. You are welcome to apply for a free personal trial version through the Sugon website ; for trials of floating licenses, cluster licenses, etc., please contact us through the Sugon website .
