Philosophy | Notion

I outline here my (Gabriel Landi) personal philosophy concerning melt.

Code should be simple

We should be able to understand the functions we are using.

I learn a lot of physics by understanding the functions I am implementing.
We usually spend more time debugging our code and doubting our results, then actually using it. Simplicity makes our lives easier.
If a 50% speed bump requires you to double the sophistication of the code, is it worth it? Sometimes, but not always.

Know your blackbox

Most Input > Output functions can be viewed as blackboxes: we don’t care what is inside, as long as it spits out the correct answer.

Not all blackboxes are equivalent, however. Specially when physics is involved. Sometimes it is very useful to use a blackbox. Sometimes it is not a good idea at all.

Here are two examples:

Eigenvalues/Eigenvectors	Quantum master equation solver
Well defined Mathematical operation	Different ways of doing it (solve an ODE, apply matrix exponential, Trotterization, &c)
Uses extremely sophisticated methods to be efficient.	Different solutions already use well-established methods (ODE, MatrixExp, &c.)
We do not necessarily learn much from understanding the algorithm.	We learn a lot from understanding the solution. Rich in physics.
✅ Use a blackbox!	❌ Do not use a blackbox! Read the code!

Melt is based on simple implementations of basic functions, which can be understood by anyone working in the field.

What is the cost?

Simplicity, of course, comes at the expense of two things: Speed and Generality.

Take QuTIP (which I am a huge fan of) as an example. Its source code is very hard to read. But it can be super fast and super general.

About speed:

Most simulations we do nowadays take on only a few minutes on a laptop. Speed is not a concern.
- In some cases it is though. If that is the case, Melt may not the appropriate choice for your problem.