Dirichlet solver

Let's play a game.

Rules

We'll play with squares. We'll call them Dirichlet squares.

They respect one, and only one rule:

Each cell contains the mean of the 4 surrounding cells.

like in:

  8
4 6 9
  3

Here: (3 + 4 + 8 + 9) / 4 == 6.

Got the rule? Let's look at the following 3×3 square:

  2 5 8
0 2 4 6 9
0 2 3 3 0
0 3 3 3 0
  7 3 6

(Yes, I added numbers outside of the square in order to check every cell in the 3×3 square complies with the rule.)

The cells can only contain positive integers (0 is allowed). The values must be exact and not rounded.

The game

So what's the game? The game is, I'll give you squares with holes (like a Sudoku) and you'll have to fill them:

  7 9 9
0       9
9       8
0       4
  6 0 9

(I'll always provide the whole border an nothing inside the actual square.)

The exercise

You'll provide a dirichlet_square_solver function, taking a single argument: a numpy array. You function will modify the array in place, filling the blanks (materialized as -1), and return nothing.

For the last example here's what I expect:

import numpy as np

square = np.array(
    [[0,  7,  9,  9, 0],
     [0, -1, -1, -1, 9],
     [9, -1, -1, -1, 8],
     [0, -1, -1, -1, 4],
     [0,  6,  0,  9, 0]],
    dtype=int)
dirichlet_square_solver(square)
print(square)
# Should give:
# [[0 7 9 9 0]
#  [0 5 7 8 9]
#  [9 6 6 7 8]
#  [0 4 4 6 4]
#  [0 6 0 9 0]]

Bonus

To do on your computer when you have the solver:

Build a huge numpy array, like 2000×6000, keep all values at 0 except for some big "hot" points or strips around it.

Solve it...

Display it using matplotlib.imshow, choose the color palette with blue for values around 0 and red for big values.

It should remind you a metal plate you heat up on some places, and it's the representation of the heat propagation in the plate.

For me, when "heating" small sections at the top of a 600×800 grid, it looks like this.