JuMP / SCS tutorial example fails on simple(?) data

Specific Domains Optimization (Mathematical)
1

Using the code verbatim as provided in the JuMP conic programming tutorial the SCS solver fails after 100,000 iterations on the matrix of five 2-D points below

julia> S
5×2 Matrix{Float64}:
 39.1  18.7
 39.5  17.4
 40.3  18.0
 36.7  19.3
 39.3  20.6

with

100000| 1.73e+14  6.42e+07  1.65e+14 -8.27e+13  1.00e-06  1.07e+00 
------------------------------------------------------------------
status:  unbounded (inaccurate - reached max_iters)
timings: total: 1.07e+00s = setup: 1.40e-04s + solve: 1.07e+00s
	 lin-sys: 8.45e-02s, cones: 8.13e-01s, accel: 2.53e-02s
------------------------------------------------------------------
objective = -inf (inaccurate)
------------------------------------------------------------------
Test Failed at /home/healyp/res/ellipse/src/ell.jl:33
  Expression: is_solved_and_feasible(model)

Could anybody please offer any advice / suggestions?

2

Try CSDP Optimizer; SCS has numerical problems with this example of S

3

Try Clarabel.jl. I’ll update the docs to use it too.

1 Like
4

Thank you – and please don’t mistake silence for ingratitude :slight_smile:

For this, and a more realistic example, CSDP ran to completion.

@odow Thank you, also, but Clarabel.jl also choked on the toy example I provided above.

1 Like
5

I am unable to reproduce this error in Clarabel.jl. The problem solves fine for me using your toy example data in all recent solver versions and on both x86_64 and aarch64.

If you are able to reproduce it, could you please open it as an issue on the solver github page?

1 Like
6

Thank you. I have just filed an issue supplying data and code.