Below is the smallest non-TU instance to the integral OHCP I have found:

Here is the Hasse diagram for :

Homology groups over :
Code output:
n_0 = 9
n_1 = 27
n_2 = 27
n_3 = 9
Dimension: 3
Euler Char: 1
c = 1*[0,1] + 1*[1,4] + 1*[4,5] + -1*[2,5] + 1*[2,3] + -1*[0,3]; lambda=0
z*=3: xp[2]=1 xm[13]=1 xp[23]=1 yp[0]=1 yp[2]=1 ym[4]=1 ym[5]=1 ym[10]=1 ym[13]=1 yp[14]=1
x = 1*[3,4] + -1*[2,4] + 1*[2,3]
y = 1*[3,4,6] + 1*[0,3,6] + -1*[1,4,6] + -1*[0,1,6] + -1*[2,4,7] + -1*[4,5,7] + 1*[2,5,7]
Every instance had . My conjecture is that this is due to the lack of torsion in the homology groups. Note that
We can get an iteration with with the following input:
54: [Feasible][P2][det(B)=2][z=4][E=[3,4,6];L=[0,3,6]] 0*[0,5] 0*[3,4,6] 0*[0,3,4] 1*[0,6] 0*[4,5,7] 0*[0,3,8] 0*[1,6,7] 0*[0,1,4] 0*[1,2,4] 1*[6,7] 0*[2,4,7] 0*[1,2] 0*[1,2,7] 0*[4,6,7] 0*[0,1] 0*[2,5,7] 0*[4,5] 0*[2,7,8] 1*[7,8] 0*[3,5,8] 0*[3,6] 0*[2,3,8] 0*[5,7,8] 0*[2,3] 1*[0,8] 0*[0,5,8] 0*[6,8]
x = 1*[0,6] + 1*[6,7] + 1*[7,8] + -1*[0,8]; y =
Computing the SNF for this instance gives us
The last row of left-transform matrix is
Under the congruence relation from 2026-05-14, this basis has integral if
That is,
for some . None of these edges are in the input vector , i.e., . Adding edge to , results in our basis no longer being encountered in the simplex method. Indeed, computing shows that adding makes the basis no longer feasible. Simply adding one of these edges to our input results in no longer being feasible.
Let be the list of edges in the above sum and consider the input vector
Using sympy, we get that the following system of inequalities for feasibility:
which sympy was unable to reduce. Adding in the constraint
gives us an ILP, which we can use software to determine if there is a solution! Solving this ILP using Google’s OR-Tools finds the following vector:
Testing finds this vector gives and . However, upon running the above vector through the OHCP LP, our basis is not encountered and we still get an integral solution. My guess, is that its because which my code assumes to solve, i.e., is actually not feasible. If we also require that , then there is no feasible solution to the ILP.