Scientific Computing Expert: Mathematics PhD ($300/Approved Task)

What you'll do

You write maths problems that an AI agent has to solve by programming. Each task is a self-contained terminal environment with data, equations, model definitions and constraints, and the agent must formulate the problem, implement a reliable algorithm, debug it and return a verifiable numerical result. The tasks belong to Terminal Bench Science.

The grading is the hard part of this variant. The ad stresses:

• Rigorous criteria covering numerical accuracy, convergence, complexity, feasibility and mathematical correctness
• Sensible tolerances, stopping criteria, stability requirements and reproducibility controls
• Tests that accept alternative valid implementations and edge cases
• Troubleshooting floating-point precision, solver failures, conditioning and performance
• Documenting assumptions, formulations, expected outputs and known limits

Topic areas named: optimisation, numerical integration, differential equations, matrix computation, statistical inference, stochastic modelling and algorithm analysis.

Who fits

• PhD, postdoctoral or equivalent advanced experience in mathematics, statistics or a close discipline
• Strong programming in Python, R, Julia, C/C++, Bash or similar
• Experience in Linux or terminal environments
• Background in numerical methods, optimisation, statistics, mathematical modelling or scientific computation
• Able to implement, test and validate algorithms on your own

What it pays

$300 per approved task, per the ad. The same figure appears on the Earth, Engineering, Life Science and Physical Sciences siblings. Only approved tasks are paid, and no quota is stated.

Worth knowing

Good:

• Published rate and a narrow, well-defined skill set
• The work rewards careful numerical thinking, which suits mathematicians who also code

Less good:

• A task that cannot be graded unambiguously may be rejected, and a rejection earns nothing
• At least 6 hours a day and 40 hours a week with 4 hours of Pacific Time overlap
• 5-week contract, no medical or paid leave
• Assessment steps are not described in the ad; Turing's help centre mentions skill tests and sometimes live interviews

About this listing

Posted by Turing on 25 September 2026 and read on 18 October 2026. Rate and terms are Turing's own; the discussion of grading difficulty is ours. See Turing.