The required minimum is 1+⌈(n+1)/5⌉ and is achieved by the configuration described in the second block of the proof.
Counting multiplicities, the cells a and c are both covered by three of these special rectangular grid arrays, the cells b and d are both covered by two, and all other red cells are covered by at least one. Letting r[xy] denote the number of red cells in [xy], it follows that r[ab]+r[bc]+r[cd]+r[da]+r[ac]≥3⋅2+2⋅2+(n−4)=n+6. Consequently, N≥(n+6)/5.
We now describe a configuration of exactly n red cells where N=1+⌈(n+1)/5⌉. Write m=⌈(n+1)/5⌉, so n=5m−r for some positive integer r≤5, and N=m+1.
Fix an integer k>2m, let S be a 3k×3k grid square, and subdivide S into nine k×k grid subsquares.
Let SLL be the lower-left corner k×k grid subsquare of S. Colour red the first m cells along the diagonal upward from the lower-right corner cell of SLL.
The ‘min’ and ‘max’ in the next four paragraphs account for the first few cases where m<r. Had we assumed n≥20, it would then have followed that m≥r, and ‘min’ and ‘max’ would have been superfluous.
Next, let SUL be the upper-left corner k×k grid subsquare of S. Colour red the first min(m,4m−r) cells along the diagonal upward from the lower-left corner cell of SUL.
Let further SUR be the upper-right corner k×k grid subsquare of S. Colour red the first min(m,3m−r) cells along the diagonal downward from the upper-left corner cell of SUR.
Complete the corner tour by letting SLR be the lower-right corner k×k grid subsquare of S. Colour red the first min(m,2m−r) cells along the diagonal downward from the upper-right corner cell of SLR.
Finally, let SC be the central k×k grid subsquare of S, and colour red max(0,m−r) cells of SC; their exact location is irrelevant.
No other cell whatsoever is coloured red, and it is a routine exercise to check that exactly n cells of the grid paper have been coloured red. Notice that, for each pair of 'adjacent' corner k×k grid subsquares, SLL and SUL, SUL and SUR, SUR and SLR, and SLR and SLL, there are both horizontal and vertical grid lines separating the strings of red cells they contain.
To complete the argument, we show that, if x and y are red cells in this configuration, then r[xy]≤m+1. This is clearly the case if x and y both lie in one of SLL, SUL, SUR, SLR or SC, for each of these squares contains at most m red cells.
If x and y lie in 'adjacent' corner k×k subsquares of S, then the red cells in [xy] come from those subsquares alone. In addition, the string of red cells in one of those subsquares has exactly one cell in [xy], namely, x or y. Consequently, r[xy]≤m+1. Incidentally, notice that equality holds if, for instance, x is the lower-right corner cell of SLL, and y is any red cell in SUL; since n≥2, there is at least one such.
If x and y lie in 'opposite' corner k×k subsquares of S, then they are the only red cells [xy] contains from those subsquares. No red cell in the other two 'opposite' corner k×k subsquares of S lies in [xy], and the other red cells in [xy] all come from SC which contains at most m−1 such. Consequently, r[xy]≤2+(m−1)=m+1.
Finally, if one of x, y lies in SC, and the other lies in one of the corner k×k sub-squares of S, then the latter cell is the only red cell in [xy] outside SC. Consequently, r[xy]≤(m−1)+1=m<m+1. This ends the proof.