Lesson Learned:

Scientific Computing Is Hard

• Limited Code Re-use
  • Libraries often do too much or too little
  • Poor documentation
  • Sometimes not much prior work available
  • Efficiency concerns
  • $$$
    
    
• Difficult Problem Domain
  • Domain expert and software developer must communicate
  • Considerable background essential
  • Correctness essential
    
    
• Small Science Changes = Large Software Changes
  • Programming language distant from problem domain
  • Testing/debugging/abandoning is costly
    

[Up to The Design and Construction of Radiative Transfer]

[Back to slide 11]