search

cvxgrp / scs

Splitting Conic Solver
MIT License
528 stars 133 forks source link

Unexpected 'Infeasible' Results with SCS 3.2.4 on QP Problems for MPC #275

Open kkzhang3 opened 1 week ago

kkzhang3 commented 1 week ago

Hi, I am using SCS 3.2.4 with Python 3.12 on a Windows 11 machine to solve Quadratic Programming (QP) problems for Model Predictive Control (MPC). Most of the optimization results are correct, and the speed is very impressive. However, I encountered three instances where the solver returned 'infeasible' results unexpectedly.

Details: SCS Version: 3.2.4 Python Version: 3.12 Operating System: Windows 11

To verify the results, I also solved the same problems using CLARABEL and CVXOPT, both of which returned 'optimal' results. I have attached the three problems that resulted in 'infeasible' status with SCS for your reference.

QPerr.zip

I turned on the export file options and get the files: data_and_logs.zip

Thanks Regards, KK Zhang

bodono commented 1 week ago

Thanks for posting this. For some reason when I run your python code I actually get 'solved' correctly form SCS (I tuned on verbose outputs from SCS to see what was happening):

=============SCS_DynamicLayer_20240625_124826_169446.pickle=============
------------------------------------------------------------------
           SCS v3.2.4 - Splitting Conic Solver
    (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07
      alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
      max_iters: 100000, normalize: 1, rho_x: 1.00e-06
      acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
      nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.27e+02  9.56e+04  2.61e+06 -1.92e+04  1.00e-01  5.62e-04
   125| 8.56e-07  7.87e-07  6.64e-10 -4.74e-01  2.60e-06  1.12e-03
------------------------------------------------------------------
status:  solved
timings: total: 1.36e-03s = setup: 4.88e-04s + solve: 8.76e-04s
     lin-sys: 3.98e-04s, cones: 1.98e-05s, accel: 9.42e-06s
------------------------------------------------------------------
objective = -0.473958
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = -0.4739577941554467, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([2.79051355e-03, 2.51015653e-03, 2.28324465e-03, 2.06283283e-03,
       1.86450968e-03, 1.68458892e-03, 1.51852915e-03, 1.39307046e-03,
       1.25959954e-03, 6.36925603e-04, 2.20305700e-03, 2.55722225e-03,
       2.01489587e-03, 1.11603498e-03, 8.82347315e-13, 2.44155017e-13,
       2.44155052e-13, 2.44154819e-13, 2.44154573e-13, 2.44154880e-13,
       2.44154818e-13, 2.44154854e-13, 2.44154832e-13, 2.44154613e-13,
       2.43690456e-13, 2.42395011e-13, 2.39947370e-13, 2.32628320e-13,
       2.28770011e-13, 2.25283848e-13, 2.19736734e-13, 2.17522596e-13,
       2.15555946e-13, 2.13101029e-13, 2.12847980e-13, 2.12966041e-13,
       2.13656973e-13, 2.13847931e-13, 2.13606401e-13, 2.12391170e-13,
       2.11936694e-13, 2.11449525e-13, 2.09675861e-13, 2.08817582e-13,
       2.08428513e-13])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.47395779415549794, clarabel_solution.status = Solved, clarabel_solution.x = [0.0027904978965449666, 0.002510198944930076, 0.0022831938027705117, 0.0020628662164093115, 0.0018645011729976001, 0.0016845818065795019, 0.001518539175488385, 0.0013930633137464905, 0.0012596049961827414, 0.0006369214890962764, 0.0022030588661007553, 0.0025572221591200366, 0.0020148959955852168, 0.0011160350187950417, 7.508467158983145e-12, 5.098200436696795e-11, 5.0919837101523314e-11, 5.086396393312051e-11, 5.076849130648014e-11, 5.0727771773585275e-11, 5.0691087613373826e-11, 5.06282096624894e-11, 5.060131621510247e-11, 5.057704866048211e-11, 5.069986578401336e-11, 5.116678636363365e-11, 5.302258154382985e-11, 5.873031052501555e-11, 6.015784957736915e-11, 6.028859646361538e-11, 5.858461187983256e-11, 5.749518217098852e-11, 5.6554606511962526e-11, 5.5554309438862856e-11, 5.550824040336608e-11, 5.5663838781453607e-11, 5.622236098014281e-11, 5.650877287399536e-11, 5.67251738200511e-11, 5.689231675566984e-11, 5.6871853878784136e-11, 5.681457007507406e-11, 5.6676452814081476e-11, 5.6626453715519324e-11, 5.659015006674711e-11]
=============SCS_DynamicLayer_20240625_124826_273466.pickle=============
------------------------------------------------------------------
           SCS v3.2.4 - Splitting Conic Solver
    (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07
      alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
      max_iters: 100000, normalize: 1, rho_x: 1.00e-06
      acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
      nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.27e+02  9.55e+04  2.61e+06 -1.87e+04  1.00e-01  4.21e-04
   125| 1.13e-06  5.85e-07  1.65e-10 -2.93e-02  1.67e-06  9.57e-04
------------------------------------------------------------------
status:  solved
timings: total: 9.61e-04s = setup: 3.57e-04s + solve: 6.05e-04s
     lin-sys: 3.97e-04s, cones: 2.77e-05s, accel: 9.54e-06s
------------------------------------------------------------------
objective = -0.029337
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = -0.02933683070721473, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([6.94261027e-04, 6.24498710e-04, 5.68067204e-04, 5.13205792e-04,
       4.63877735e-04, 4.19117973e-04, 3.77790668e-04, 3.46590440e-04,
       3.13374278e-04, 1.58465232e-04, 5.48101854e-04, 6.36216877e-04,
       5.01290169e-04, 2.77660710e-04, 8.97944892e-13, 2.44076024e-13,
       2.44076057e-13, 2.44075891e-13, 2.44076235e-13, 2.44076090e-13,
       2.44076115e-13, 2.44076194e-13, 2.44076076e-13, 2.44076086e-13,
       2.43611958e-13, 2.42316618e-13, 2.39869035e-13, 2.32552681e-13,
       2.28695071e-13, 2.25209971e-13, 2.19664083e-13, 2.17450745e-13,
       2.15484662e-13, 2.13030405e-13, 2.12777627e-13, 2.12895381e-13,
       2.13586682e-13, 2.13777081e-13, 2.13535797e-13, 2.12320797e-13,
       2.11866598e-13, 2.11379410e-13, 2.09606049e-13, 2.08748222e-13,
       2.08359357e-13])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.029336830707231296, clarabel_solution.status = Solved, clarabel_solution.x = [0.0006942523340928437, 0.0006245211531611647, 0.0005680402154852255, 0.0005132247970703782, 0.0004638732525990768, 0.00041911069714191607, 0.0003778005965243507, 0.0003465831910797031, 0.00031337981165299804, 0.00015846105492833785, 0.0005481037104770503, 0.0006362167512226497, 0.0005012902690682797, 0.00027766073090962665, 7.422584398118871e-12, 4.9908818125428526e-11, 4.989439411164203e-11, 4.988135885249981e-11, 4.985892511625296e-11, 4.9849294577522e-11, 4.9840585958786774e-11, 4.982558652410234e-11, 4.9819142820566604e-11, 4.981331356680122e-11, 4.996139585601154e-11, 5.043061010730071e-11, 5.2267431264695483e-11, 5.791302606624232e-11, 5.933302998604015e-11, 5.947440433414254e-11, 5.7813882893553324e-11, 5.674740148239301e-11, 5.582688709371417e-11, 5.485106791023405e-11, 5.481192913018784e-11, 5.497105449305245e-11, 5.55308513300546e-11, 5.5818938404726124e-11, 5.603741130163767e-11, 5.620771956517265e-11, 5.618918130039591e-11, 5.61345389366892e-11, 5.6001086068957654e-11, 5.595235055766623e-11, 5.5916314191366856e-11]
=============SCS_DynamicLayer_20240625_124826_366924.pickle=============
------------------------------------------------------------------
           SCS v3.2.4 - Splitting Conic Solver
    (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07
      alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
      max_iters: 100000, normalize: 1, rho_x: 1.00e-06
      acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
      nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.26e+02  9.55e+04  2.61e+06 -1.86e+04  1.00e-01  4.15e-04
   125| 1.18e-06  4.87e-07  1.03e-10 -2.22e-03  1.08e-06  9.46e-04
------------------------------------------------------------------
status:  solved
timings: total: 9.54e-04s = setup: 3.51e-04s + solve: 6.03e-04s
     lin-sys: 4.00e-04s, cones: 1.99e-05s, accel: 9.37e-06s
------------------------------------------------------------------
objective = -0.002215
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = -0.002215217426750132, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([1.90769690e-04, 1.71594496e-04, 1.56110165e-04, 1.41011648e-04,
       1.27465265e-04, 1.15173374e-04, 1.03803298e-04, 9.52430020e-05,
       8.61068673e-05, 4.35357962e-05, 1.50626706e-04, 1.74800288e-04,
       1.37827614e-04, 7.63372245e-05, 9.03386108e-13, 2.44050817e-13,
       2.44050751e-13, 2.44050780e-13, 2.44050750e-13, 2.44050759e-13,
       2.44050816e-13, 2.44050761e-13, 2.44050817e-13, 2.44050791e-13,
       2.43586577e-13, 2.42291618e-13, 2.39844182e-13, 2.32528051e-13,
       2.28670874e-13, 2.25186067e-13, 2.19640423e-13, 2.17427416e-13,
       2.15461294e-13, 2.13007261e-13, 2.12754315e-13, 2.12872044e-13,
       2.13563177e-13, 2.13753984e-13, 2.13512252e-13, 2.12297703e-13,
       2.11843459e-13, 2.11356310e-13, 2.09583286e-13, 2.08725420e-13,
       2.08336267e-13])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.0022152174267624503, clarabel_solution.status = Solved, clarabel_solution.x = [0.00019076221443803517, 0.00017161401041454087, 0.00015608744276640255, 0.00014102549607533437, 0.00012746475443084693, 0.00011516469875157536, 0.00010381329027656681, 9.523573477537405e-05, 8.611244900019338e-05, 4.353157804871756e-05, 0.0001506285695843726, 0.00017480015346372276, 0.0001378277079923815, 7.633724223309769e-05, 7.402155371667622e-12, 4.965827303819483e-11, 4.965437612207606e-11, 4.9650849634359917e-11, 4.9644769875527325e-11, 4.9642155753222594e-11, 4.963978972099507e-11, 4.963570972671434e-11, 4.9633955095898954e-11, 4.963236680374209e-11, 4.9786226003858936e-11, 5.025590754635561e-11, 5.2088158356424624e-11, 5.771888610298373e-11, 5.913705369744592e-11, 5.928091578082756e-11, 5.763067279392684e-11, 5.6569626886655146e-11, 5.565386489883837e-11, 5.4683841907138966e-11, 5.464633867216354e-11, 5.48062925637533e-11, 5.5366378165935806e-11, 5.5654855288171645e-11, 5.587381381126671e-11, 5.6044868611528313e-11, 5.602678592850865e-11, 5.5972769367126994e-11, 5.584042230149203e-11, 5.5791986607535523e-11, 5.575601417531867e-11]
bodono commented 1 week ago

Can you post the output of what you get when you run with verbose=True? It might be that something is compiled wrongly in your setup somehow.

kkzhang3 commented 1 week ago

Sure. These are the results.

scs.__version__ = '3.2.4'
clarabel.__version__ = '0.7.1'
=============SCS_DynamicLayer_20240625_124826_169446.pickle=============
------------------------------------------------------------------      
               SCS v3.2.4 - Splitting Conic Solver
        (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------      
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07       
          alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
          max_iters: 100000, normalize: 1, rho_x: 1.00e-06
          acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
          nnz(A): 770, nnz(P): 571
------------------------------------------------------------------      
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.27e+02  9.56e+04  2.61e+06 -1.92e+04  1.00e-01  1.50e-03
    25| 1.07e-05  1.73e+01  1.92e+01 -4.92e-01  1.00e-01  1.78e-03
------------------------------------------------------------------
status:  infeasible
timings: total: 1.90e-03s = setup: 9.41e-04s + solve: 9.60e-04s
         lin-sys: 1.26e-04s, cones: 9.50e-06s, accel: 1.40e-06s
------------------------------------------------------------------
objective = inf
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = inf, SCSsol['info']['status'] = 'infeasible', SCSsol['x'] = array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.47395779415549794, clarabel_solution.status = Solved, clarabel_solution.x = [0.0027904978965449666, 0.002510198944930076, 0.0022831938027705117, 0.0020628662164093115, 0.0018645011729976001, 0.0016845818065795019, 0.001518539175488385, 0.0013930633137464905, 0.0012596049961827414, 0.0006369214890962764, 0.0022030588661007553, 0.0025572221591200366, 0.0020148959955852168, 0.0011160350187950417, 7.508467158983145e-12, 5.098200436696795e-11, 5.0919837101523314e-11, 5.086396393312051e-11, 5.076849130648014e-11, 
5.0727771773585275e-11, 5.0691087613373826e-11, 5.06282096624894e-11, 5.060131621510247e-11, 5.057704866048211e-11, 5.069986578401336e-11, 5.116678636363365e-11, 5.302258154382985e-11, 5.873031052501555e-11, 6.015784957736915e-11, 6.028859646361538e-11, 5.858461187983256e-11, 5.749518217098852e-11, 5.6554606511962526e-11, 5.5554309438862856e-11, 5.550824040336608e-11, 5.5663838781453607e-11, 5.622236098014281e-11, 5.650877287399536e-11, 5.67251738200511e-11, 5.689231675566984e-11, 5.6871853878784136e-11, 5.681457007507406e-11, 5.6676452814081476e-11, 5.6626453715519324e-11, 5.659015006674711e-11]
=============SCS_DynamicLayer_20240625_124826_273466.pickle=============
------------------------------------------------------------------
               SCS v3.2.4 - Splitting Conic Solver
        (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07
          alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
          max_iters: 100000, normalize: 1, rho_x: 1.00e-06
          acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
          nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.27e+02  9.55e+04  2.61e+06 -1.87e+04  1.00e-01  2.44e-03
    75| 1.59e-13  1.45e+00  2.50e-01 -2.95e-02  1.00e-01  2.98e-03
------------------------------------------------------------------
status:  infeasible
timings: total: 3.11e-03s = setup: 2.00e-03s + solve: 1.11e-03s
         lin-sys: 3.29e-04s, cones: 2.03e-05s, accel: 5.90e-06s
------------------------------------------------------------------
objective = inf
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = inf, SCSsol['info']['status'] = 'infeasible', SCSsol['x'] = array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.029336830707231296, clarabel_solution.status = Solved, clarabel_solution.x = [0.0006942523340928437, 0.0006245211531611647, 0.0005680402154852255, 0.0005132247970703782, 0.0004638732525990768, 0.00041911069714191607, 0.0003778005965243507, 0.0003465831910797031, 0.00031337981165299804, 0.00015846105492833785, 0.0005481037104770503, 0.0006362167512226497, 0.0005012902690682797, 0.00027766073090962665, 7.422584398118871e-12, 4.9908818125428526e-11, 4.989439411164203e-11, 4.988135885249981e-11, 4.985892511625296e-11, 4.9849294577522e-11, 4.9840585958786774e-11, 4.982558652410234e-11, 4.9819142820566604e-11, 4.981331356680122e-11, 4.996139585601154e-11, 5.043061010730071e-11, 5.2267431264695483e-11, 5.791302606624232e-11, 5.933302998604015e-11, 5.947440433414254e-11, 5.7813882893553324e-11, 5.674740148239301e-11, 5.582688709371417e-11, 5.485106791023405e-11, 5.481192913018784e-11, 5.497105449305245e-11, 5.55308513300546e-11, 5.5818938404726124e-11, 5.603741130163767e-11, 5.620771956517265e-11, 5.618918130039591e-11, 5.61345389366892e-11, 5.6001086068957654e-11, 5.595235055766623e-11, 5.5916314191366856e-11]
=============SCS_DynamicLayer_20240625_124826_366924.pickle=============
------------------------------------------------------------------
               SCS v3.2.4 - Splitting Conic Solver
        (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07
          alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
          max_iters: 100000, normalize: 1, rho_x: 1.00e-06
          acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
          nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.26e+02  9.55e+04  2.61e+06 -1.86e+04  1.00e-01  2.09e-03
    50| 3.46e-13  4.37e+00  1.68e+00 -2.37e-03  1.00e-01  2.46e-03
------------------------------------------------------------------
status:  infeasible
timings: total: 2.57e-03s = setup: 1.69e-03s + solve: 8.78e-04s
         lin-sys: 2.05e-04s, cones: 1.33e-05s, accel: 3.60e-06s
------------------------------------------------------------------
objective = inf
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = inf, SCSsol['info']['status'] = 'infeasible', SCSsol['x'] = array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.0022152174267624503, clarabel_solution.status = Solved, clarabel_solution.x = [0.00019076221443803517, 0.00017161401041454087, 0.00015608744276640255, 0.00014102549607533437, 0.00012746475443084693, 0.00011516469875157536, 0.00010381329027656681, 9.523573477537405e-05, 8.611244900019338e-05, 4.353157804871756e-05, 0.0001506285695843726, 0.00017480015346372276, 0.0001378277079923815, 7.633724223309769e-05, 7.402155371667622e-12, 4.965827303819483e-11, 4.965437612207606e-11, 4.9650849634359917e-11, 4.9644769875527325e-11, 4.9642155753222594e-11, 4.963978972099507e-11, 4.963570972671434e-11, 4.9633955095898954e-11, 4.963236680374209e-11, 4.9786226003858936e-11, 5.025590754635561e-11, 5.2088158356424624e-11, 5.771888610298373e-11, 5.913705369744592e-11, 5.928091578082756e-11, 5.763067279392684e-11, 5.6569626886655146e-11, 5.565386489883837e-11, 5.4683841907138966e-11, 5.464633867216354e-11, 5.48062925637533e-11, 5.5366378165935806e-11, 5.5654855288171645e-11, 5.587381381126671e-11, 5.6044868611528313e-11, 5.602678592850865e-11, 5.5972769367126994e-11, 5.584042230149203e-11, 5.5791986607535523e-11, 5.575601417531867e-11]
bodono commented 1 week ago

Thanks for posting this! I still have not been able to reproduce this but I think I have identified the problem, a very nasty bug in how I'm calling the Blas dlange method, for calculating the infinity norm. I'm amazed this didn't trigger a bug before!

If you are able to test please can you patch in this change: https://github.com/cvxgrp/scs/pull/278/commits/8891c9cd4e9d914acecaef3846133f9bde1cf030

And then retry? I can talk you through how to do this if you're having trouble.

ymwang78 commented 1 week ago

build source from main branch (the change is not applied), seems no problem:

scs.version = '3.2.4' clarabel.version = '0.9.0' =============SCS_DynamicLayer_20240625_124826_169446.pickle============= -----------SCS----------- SCSsol['info']['pobj'] = -0.47395779415545064, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([2.79051286e-03, 2.51015842e-03, 2.28324254e-03, 2.06283394e-03, 1.86450946e-03, 1.68458889e-03, 1.51852915e-03, 1.39307048e-03, 1.25959953e-03, 6.36925612e-04, 2.20305700e-03, 2.55722225e-03, 2.01489587e-03, 1.11603498e-03, 8.84477770e-13, 2.44157244e-13, 2.44157344e-13, 2.44157046e-13, 2.44157365e-13, 2.44157298e-13, 2.44156987e-13, 2.44157257e-13, 2.44157240e-13, 2.44157125e-13, 2.43693110e-13, 2.42397618e-13, 2.39949448e-13, 2.32630732e-13, 2.28772681e-13, 2.25286026e-13, 2.19738649e-13, 2.17524842e-13, 2.15558016e-13, 2.13102962e-13, 2.12850018e-13, 2.12967835e-13, 2.13659155e-13, 2.13849878e-13, 2.13608392e-13, 2.12393270e-13, 2.11938506e-13, 2.11451457e-13, 2.09677605e-13, 2.08819459e-13, 2.08430287e-13]) -----------CLARABEL----------- clarabel_solution.obj_val = -0.47395779415549794, clarabel_solution.status = Solved, clarabel_solution.x = [0.0027904978965449666, 0.002510198944930076, 0.0022831938027705117, 0.0020628662164093115, 0.0018645011729976001, 0.0016845818065795019, 0.001518539175488385, 0.0013930633137464905, 0.0012596049961827414, 0.0006369214890962764, 0.0022030588661007553, 0.0025572221591200366, 0.0020148959955852168, 0.0011160350187950417, 7.508467158983145e-12, 5.098200436696795e-11, 5.0919837101523314e-11, 5.086396393312051e-11, 5.076849130648014e-11, 5.0727771773585275e-11, 5.0691087613373826e-11, 5.06282096624894e-11, 5.060131621510247e-11, 5.057704866048211e-11, 5.069986578401336e-11, 5.116678636363365e-11, 5.302258154382985e-11, 5.873031052501555e-11, 6.015784957736915e-11, 6.028859646361538e-11, 5.858461187983256e-11, 5.749518217098852e-11, 5.6554606511962526e-11, 5.5554309438862856e-11, 5.550824040336608e-11, 5.5663838781453607e-11, 5.622236098014281e-11, 5.650877287399536e-11, 5.67251738200511e-11, 5.689231675566984e-11, 5.6871853878784136e-11, 5.681457007507406e-11, 5.6676452814081476e-11, 5.6626453715519324e-11, 5.659015006674711e-11] =============SCS_DynamicLayer_20240625_124826_273466.pickle============= -----------SCS----------- SCSsol['info']['pobj'] = -0.029336830707218327, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([6.94259936e-04, 6.24501827e-04, 5.68063419e-04, 5.13208238e-04, 4.63876894e-04, 4.19118079e-04, 3.77790690e-04, 3.46590442e-04, 3.13374274e-04, 1.58465238e-04, 5.48101857e-04, 6.36216882e-04, 5.01290173e-04, 2.77660712e-04, 8.99178393e-13, 2.44076312e-13, 2.44076201e-13, 2.44076130e-13, 2.44076233e-13, 2.44076309e-13, 2.44076333e-13, 2.44076175e-13, 2.44076303e-13, 2.44076117e-13, 2.43612146e-13, 2.42317150e-13, 2.39869209e-13, 2.32552922e-13, 2.28695235e-13, 2.25210428e-13, 2.19664045e-13, 2.17451237e-13, 2.15484886e-13, 2.13030754e-13, 2.12777663e-13, 2.12895197e-13, 2.13586552e-13, 2.13777236e-13, 2.13535904e-13, 2.12321086e-13, 2.11866553e-13, 2.11379857e-13, 2.09606355e-13, 2.08748669e-13, 2.08359173e-13]) -----------CLARABEL----------- clarabel_solution.obj_val = -0.029336830707231296, clarabel_solution.status = Solved, clarabel_solution.x = [0.0006942523340928437, 0.0006245211531611647, 0.0005680402154852255, 0.0005132247970703782, 0.0004638732525990768, 0.00041911069714191607, 0.0003778005965243507, 0.0003465831910797031, 0.00031337981165299804, 0.00015846105492833785, 0.0005481037104770503, 0.0006362167512226497, 0.0005012902690682797, 0.00027766073090962665, 7.422584398118871e-12, 4.9908818125428526e-11, 4.989439411164203e-11, 4.988135885249981e-11, 4.985892511625296e-11, 4.9849294577522e-11, 4.9840585958786774e-11, 4.982558652410234e-11, 4.9819142820566604e-11, 4.981331356680122e-11, 4.996139585601154e-11, 5.043061010730071e-11, 5.2267431264695483e-11, 5.791302606624232e-11, 5.933302998604015e-11, 5.947440433414254e-11, 5.7813882893553324e-11, 5.674740148239301e-11, 5.582688709371417e-11, 5.485106791023405e-11, 5.481192913018784e-11, 5.497105449305245e-11, 5.55308513300546e-11, 5.5818938404726124e-11, 5.603741130163767e-11, 5.620771956517265e-11, 5.618918130039591e-11, 5.61345389366892e-11, 5.6001086068957654e-11, 5.595235055766623e-11, 5.5916314191366856e-11] =============SCS_DynamicLayer_20240625_124826_366924.pickle============= -----------SCS----------- SCSsol['info']['pobj'] = -0.002215217426752619, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([1.90768627e-04, 1.71597468e-04, 1.56107152e-04, 1.41012605e-04, 1.27465658e-04, 1.15173066e-04, 1.03803338e-04, 9.52429990e-05, 8.61068786e-05, 4.35357896e-05, 1.50626712e-04, 1.74800292e-04, 1.37827617e-04, 7.63372261e-05, 9.02991412e-13, 2.44051336e-13, 2.44051360e-13, 2.44051268e-13, 2.44051210e-13, 2.44051394e-13, 2.44051256e-13, 2.44051419e-13, 2.44051233e-13, 2.44051275e-13, 2.43587140e-13, 2.42292083e-13, 2.39844607e-13, 2.32528533e-13, 2.28671382e-13, 2.25186628e-13, 2.19641011e-13, 2.17428008e-13, 2.15461903e-13, 2.13008008e-13, 2.12755080e-13, 2.12872718e-13, 2.13563827e-13, 2.13754464e-13, 2.13512943e-13, 2.12298399e-13, 2.11844188e-13, 2.11356916e-13, 2.09583937e-13, 2.08726103e-13, 2.08336882e-13]) -----------CLARABEL----------- clarabel_solution.obj_val = -0.0022152174267624503, clarabel_solution.status = Solved, clarabel_solution.x = [0.00019076221443803517, 0.00017161401041454087, 0.00015608744276640255, 0.00014102549607533437, 0.00012746475443084693, 0.00011516469875157536, 0.00010381329027656681, 9.523573477537405e-05, 8.611244900019338e-05, 4.353157804871756e-05, 0.0001506285695843726, 0.00017480015346372276, 0.0001378277079923815, 7.633724223309769e-05, 7.402155371667622e-12, 4.965827303819483e-11, 4.965437612207606e-11, 4.9650849634359917e-11, 4.9644769875527325e-11, 4.9642155753222594e-11, 4.963978972099507e-11, 4.963570972671434e-11, 4.9633955095898954e-11, 4.963236680374209e-11, 4.9786226003858936e-11, 5.025590754635561e-11, 5.2088158356424624e-11, 5.771888610298373e-11, 5.913705369744592e-11, 5.928091578082756e-11, 5.763067279392684e-11, 5.6569626886655146e-11, 5.565386489883837e-11, 5.4683841907138966e-11, 5.464633867216354e-11, 5.48062925637533e-11, 5.5366378165935806e-11, 5.5654855288171645e-11, 5.587381381126671e-11, 5.6044868611528313e-11, 5.602678592850865e-11, 5.5972769367126994e-11, 5.584042230149203e-11, 5.5791986607535523e-11, 5.575601417531867e-11]

ymwang78 commented 1 week ago

After applied the patch, got the same result:

scs.version = '3.2.4' clarabel.version = '0.9.0' =============SCS_DynamicLayer_20240625_124826_169446.pickle============= -----------SCS----------- SCSsol['info']['pobj'] = -0.47395779415545064, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([2.79051286e-03, 2.51015842e-03, 2.28324254e-03, 2.06283394e-03, 1.86450946e-03, 1.68458889e-03, 1.51852915e-03, 1.39307048e-03, 1.25959953e-03, 6.36925612e-04, 2.20305700e-03, 2.55722225e-03, 2.01489587e-03, 1.11603498e-03, 8.84477770e-13, 2.44157244e-13, 2.44157344e-13, 2.44157046e-13, 2.44157365e-13, 2.44157298e-13, 2.44156987e-13, 2.44157257e-13, 2.44157240e-13, 2.44157125e-13, 2.43693110e-13, 2.42397618e-13, 2.39949448e-13, 2.32630732e-13, 2.28772681e-13, 2.25286026e-13, 2.19738649e-13, 2.17524842e-13, 2.15558016e-13, 2.13102962e-13, 2.12850018e-13, 2.12967835e-13, 2.13659155e-13, 2.13849878e-13, 2.13608392e-13, 2.12393270e-13, 2.11938506e-13, 2.11451457e-13, 2.09677605e-13, 2.08819459e-13, 2.08430287e-13]) -----------CLARABEL----------- clarabel_solution.obj_val = -0.47395779415549794, clarabel_solution.status = Solved, clarabel_solution.x = [0.0027904978965449666, 0.002510198944930076, 0.0022831938027705117, 0.0020628662164093115, 0.0018645011729976001, 0.0016845818065795019, 0.001518539175488385, 0.0013930633137464905, 0.0012596049961827414, 0.0006369214890962764, 0.0022030588661007553, 0.0025572221591200366, 0.0020148959955852168, 0.0011160350187950417, 7.508467158983145e-12, 5.098200436696795e-11, 5.0919837101523314e-11, 5.086396393312051e-11, 5.076849130648014e-11, 5.0727771773585275e-11, 5.0691087613373826e-11, 5.06282096624894e-11, 5.060131621510247e-11, 5.057704866048211e-11, 5.069986578401336e-11, 5.116678636363365e-11, 5.302258154382985e-11, 5.873031052501555e-11, 6.015784957736915e-11, 6.028859646361538e-11, 5.858461187983256e-11, 5.749518217098852e-11, 5.6554606511962526e-11, 5.5554309438862856e-11, 5.550824040336608e-11, 5.5663838781453607e-11, 5.622236098014281e-11, 5.650877287399536e-11, 5.67251738200511e-11, 5.689231675566984e-11, 5.6871853878784136e-11, 5.681457007507406e-11, 5.6676452814081476e-11, 5.6626453715519324e-11, 5.659015006674711e-11] =============SCS_DynamicLayer_20240625_124826_273466.pickle============= -----------SCS----------- SCSsol['info']['pobj'] = -0.029336830707218327, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([6.94259936e-04, 6.24501827e-04, 5.68063419e-04, 5.13208238e-04, 4.63876894e-04, 4.19118079e-04, 3.77790690e-04, 3.46590442e-04, 3.13374274e-04, 1.58465238e-04, 5.48101857e-04, 6.36216882e-04, 5.01290173e-04, 2.77660712e-04, 8.99178393e-13, 2.44076312e-13, 2.44076201e-13, 2.44076130e-13, 2.44076233e-13, 2.44076309e-13, 2.44076333e-13, 2.44076175e-13, 2.44076303e-13, 2.44076117e-13, 2.43612146e-13, 2.42317150e-13, 2.39869209e-13, 2.32552922e-13, 2.28695235e-13, 2.25210428e-13, 2.19664045e-13, 2.17451237e-13, 2.15484886e-13, 2.13030754e-13, 2.12777663e-13, 2.12895197e-13, 2.13586552e-13, 2.13777236e-13, 2.13535904e-13, 2.12321086e-13, 2.11866553e-13, 2.11379857e-13, 2.09606355e-13, 2.08748669e-13, 2.08359173e-13]) -----------CLARABEL----------- clarabel_solution.obj_val = -0.029336830707231296, clarabel_solution.status = Solved, clarabel_solution.x = [0.0006942523340928437, 0.0006245211531611647, 0.0005680402154852255, 0.0005132247970703782, 0.0004638732525990768, 0.00041911069714191607, 0.0003778005965243507, 0.0003465831910797031, 0.00031337981165299804, 0.00015846105492833785, 0.0005481037104770503, 0.0006362167512226497, 0.0005012902690682797, 0.00027766073090962665, 7.422584398118871e-12, 4.9908818125428526e-11, 4.989439411164203e-11, 4.988135885249981e-11, 4.985892511625296e-11, 4.9849294577522e-11, 4.9840585958786774e-11, 4.982558652410234e-11, 4.9819142820566604e-11, 4.981331356680122e-11, 4.996139585601154e-11, 5.043061010730071e-11, 5.2267431264695483e-11, 5.791302606624232e-11, 5.933302998604015e-11, 5.947440433414254e-11, 5.7813882893553324e-11, 5.674740148239301e-11, 5.582688709371417e-11, 5.485106791023405e-11, 5.481192913018784e-11, 5.497105449305245e-11, 5.55308513300546e-11, 5.5818938404726124e-11, 5.603741130163767e-11, 5.620771956517265e-11, 5.618918130039591e-11, 5.61345389366892e-11, 5.6001086068957654e-11, 5.595235055766623e-11, 5.5916314191366856e-11] =============SCS_DynamicLayer_20240625_124826_366924.pickle============= -----------SCS----------- SCSsol['info']['pobj'] = -0.002215217426752619, SCSsol['info']['status'] = 'solved', SCSsol['x'] = array([1.90768627e-04, 1.71597468e-04, 1.56107152e-04, 1.41012605e-04, 1.27465658e-04, 1.15173066e-04, 1.03803338e-04, 9.52429990e-05, 8.61068786e-05, 4.35357896e-05, 1.50626712e-04, 1.74800292e-04, 1.37827617e-04, 7.63372261e-05, 9.02991412e-13, 2.44051336e-13, 2.44051360e-13, 2.44051268e-13, 2.44051210e-13, 2.44051394e-13, 2.44051256e-13, 2.44051419e-13, 2.44051233e-13, 2.44051275e-13, 2.43587140e-13, 2.42292083e-13, 2.39844607e-13, 2.32528533e-13, 2.28671382e-13, 2.25186628e-13, 2.19641011e-13, 2.17428008e-13, 2.15461903e-13, 2.13008008e-13, 2.12755080e-13, 2.12872718e-13, 2.13563827e-13, 2.13754464e-13, 2.13512943e-13, 2.12298399e-13, 2.11844188e-13, 2.11356916e-13, 2.09583937e-13, 2.08726103e-13, 2.08336882e-13]) -----------CLARABEL----------- clarabel_solution.obj_val = -0.0022152174267624503, clarabel_solution.status = Solved, clarabel_solution.x = [0.00019076221443803517, 0.00017161401041454087, 0.00015608744276640255, 0.00014102549607533437, 0.00012746475443084693, 0.00011516469875157536, 0.00010381329027656681, 9.523573477537405e-05, 8.611244900019338e-05, 4.353157804871756e-05, 0.0001506285695843726, 0.00017480015346372276, 0.0001378277079923815, 7.633724223309769e-05, 7.402155371667622e-12, 4.965827303819483e-11, 4.965437612207606e-11, 4.9650849634359917e-11, 4.9644769875527325e-11, 4.9642155753222594e-11, 4.963978972099507e-11, 4.963570972671434e-11, 4.9633955095898954e-11, 4.963236680374209e-11, 4.9786226003858936e-11, 5.025590754635561e-11, 5.2088158356424624e-11, 5.771888610298373e-11, 5.913705369744592e-11, 5.928091578082756e-11, 5.763067279392684e-11, 5.6569626886655146e-11, 5.565386489883837e-11, 5.4683841907138966e-11, 5.464633867216354e-11, 5.48062925637533e-11, 5.5366378165935806e-11, 5.5654855288171645e-11, 5.587381381126671e-11, 5.6044868611528313e-11, 5.602678592850865e-11, 5.5972769367126994e-11, 5.584042230149203e-11, 5.5791986607535523e-11, 5.575601417531867e-11]

kkzhang3 commented 1 week ago

I asked my colleague @ymwang78, who is much more knowledgeable in C and C++ than I am, to help me with this issue. We discovered that if we install SCS through pip, we both encounter the same infeasibility error. However, when we build SCS from the original source code, we get the correct results. This observation holds true for both the main branch and the new modified branch 8891c9c.

We use python 3.12 on windows 11 system.

bodono commented 1 week ago

Thanks for all the information. I can only think of the difference when installing via pip is from linking against a different blas library, but I'm surprised that the change didn't fix it. I am planning to merge that fix and cut a new pip release with it, so it should be easy to just use pip install again to pick up the new version and we can test again then. This is very mysterious!

bodono commented 1 week ago

Ok i just cut a new pip release, can you pip install or pip upgrade and retry? Make sure it's version 3.2.5. Fingers crossed this fixes it.

kkzhang3 commented 1 week ago

I tried version 3.2.5 by installing it via pip. To my surprise, the infeasibility issue still persists. 

scs.__version__ = '3.2.5'
clarabel.__version__ = '0.7.1'
=============SCS_DynamicLayer_20240625_124826_169446.pickle=============
------------------------------------------------------------------      
               SCS v3.2.5 - Splitting Conic Solver
        (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------      
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07       
          alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
          max_iters: 100000, normalize: 1, rho_x: 1.00e-06
          acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
          nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.27e+02  9.56e+04  2.61e+06 -1.92e+04  1.00e-01  2.36e-03
    25| 1.07e-05  1.73e+01  1.92e+01 -4.92e-01  1.00e-01  2.63e-03
------------------------------------------------------------------
status:  infeasible
timings: total: 2.76e-03s = setup: 1.17e-03s + solve: 1.59e-03s
         lin-sys: 1.04e-04s, cones: 1.09e-05s, accel: 1.60e-06s
------------------------------------------------------------------
objective = inf
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = inf, SCSsol['info']['status'] = 'infeasible', SCSsol['x'] = array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.47395779415549794, clarabel_solution.status = Solved, clarabel_solution.x = [0.0027904978965449666, 0.002510198944930076, 0.0022831938027705117, 0.0020628662164093115, 0.0018645011729976001, 0.0016845818065795019, 0.001518539175488385, 0.0013930633137464905, 0.0012596049961827414, 0.0006369214890962764, 0.0022030588661007553, 0.0025572221591200366, 0.0020148959955852168, 0.0011160350187950417, 7.508467158983145e-12, 5.098200436696795e-11, 5.0919837101523314e-11, 5.086396393312051e-11, 5.076849130648014e-11, 5.0727771773585275e-11, 5.0691087613373826e-11, 5.06282096624894e-11, 5.060131621510247e-11, 5.057704866048211e-11, 
5.069986578401336e-11, 5.116678636363365e-11, 5.302258154382985e-11, 5.873031052501555e-11, 6.015784957736915e-11, 6.028859646361538e-11, 5.858461187983256e-11, 5.749518217098852e-11, 5.6554606511962526e-11, 5.5554309438862856e-11, 5.550824040336608e-11, 5.5663838781453607e-11, 5.622236098014281e-11, 5.650877287399536e-11, 5.67251738200511e-11, 5.689231675566984e-11, 5.6871853878784136e-11, 5.681457007507406e-11, 5.6676452814081476e-11, 5.6626453715519324e-11, 5.659015006674711e-11]
=============SCS_DynamicLayer_20240625_124826_273466.pickle=============
------------------------------------------------------------------
               SCS v3.2.5 - Splitting Conic Solver
        (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07
          alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
          max_iters: 100000, normalize: 1, rho_x: 1.00e-06
          acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
          nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.27e+02  9.55e+04  2.61e+06 -1.87e+04  1.00e-01  3.44e-03
    75| 1.59e-13  1.45e+00  2.50e-01 -2.95e-02  1.00e-01  4.17e-03
------------------------------------------------------------------
status:  infeasible
timings: total: 4.30e-03s = setup: 2.76e-03s + solve: 1.54e-03s
         lin-sys: 4.51e-04s, cones: 3.50e-05s, accel: 1.15e-05s
------------------------------------------------------------------
objective = inf
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = inf, SCSsol['info']['status'] = 'infeasible', SCSsol['x'] = array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.029336830707231296, clarabel_solution.status = Solved, clarabel_solution.x = [0.0006942523340928437, 0.0006245211531611647, 0.0005680402154852255, 0.0005132247970703782, 0.0004638732525990768, 0.00041911069714191607, 0.0003778005965243507, 0.0003465831910797031, 0.00031337981165299804, 0.00015846105492833785, 0.0005481037104770503, 0.0006362167512226497, 0.0005012902690682797, 0.00027766073090962665, 7.422584398118871e-12, 4.9908818125428526e-11, 4.989439411164203e-11, 4.988135885249981e-11, 4.985892511625296e-11, 4.9849294577522e-11, 4.9840585958786774e-11, 4.982558652410234e-11, 4.9819142820566604e-11, 4.981331356680122e-11, 4.996139585601154e-11, 5.043061010730071e-11, 5.2267431264695483e-11, 5.791302606624232e-11, 5.933302998604015e-11, 5.947440433414254e-11, 5.7813882893553324e-11, 5.674740148239301e-11, 5.582688709371417e-11, 5.485106791023405e-11, 5.481192913018784e-11, 5.497105449305245e-11, 5.55308513300546e-11, 5.5818938404726124e-11, 5.603741130163767e-11, 5.620771956517265e-11, 5.618918130039591e-11, 5.61345389366892e-11, 5.6001086068957654e-11, 5.595235055766623e-11, 5.5916314191366856e-11]
=============SCS_DynamicLayer_20240625_124826_366924.pickle=============
------------------------------------------------------------------
               SCS v3.2.5 - Splitting Conic Solver
        (c) Brendan O'Donoghue, Stanford University, 2012
------------------------------------------------------------------
problem:  variables n: 45, constraints m: 120
cones:    l: linear vars: 120
settings: eps_abs: 1.0e-07, eps_rel: 1.0e-06, eps_infeas: 1.0e-07
          alpha: 1.50, scale: 1.00e-01, adaptive_scale: 1
          max_iters: 100000, normalize: 1, rho_x: 1.00e-06
          acceleration_lookback: 10, acceleration_interval: 10
lin-sys:  sparse-direct-amd-qdldl
          nnz(A): 770, nnz(P): 571
------------------------------------------------------------------
 iter | pri res | dua res |   gap   |   obj   |  scale  | time (s)
------------------------------------------------------------------
     0| 2.26e+02  9.55e+04  2.61e+06 -1.86e+04  1.00e-01  3.43e-03
    50| 3.46e-13  4.37e+00  1.68e+00 -2.37e-03  1.00e-01  3.83e-03
------------------------------------------------------------------
status:  infeasible
timings: total: 3.94e-03s = setup: 2.63e-03s + solve: 1.32e-03s
         lin-sys: 2.17e-04s, cones: 1.64e-05s, accel: 5.50e-06s
------------------------------------------------------------------
objective = inf
------------------------------------------------------------------
-----------SCS-----------
SCSsol['info']['pobj'] = inf, SCSsol['info']['status'] = 'infeasible', SCSsol['x'] = array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan,
       nan, nan, nan, nan, nan, nan])
-----------CLARABEL-----------
clarabel_solution.obj_val = -0.0022152174267624503, clarabel_solution.status = Solved, clarabel_solution.x = [0.00019076221443803517, 0.00017161401041454087, 0.00015608744276640255, 0.00014102549607533437, 0.00012746475443084693, 0.00011516469875157536, 0.00010381329027656681, 9.523573477537405e-05, 8.611244900019338e-05, 4.353157804871756e-05, 0.0001506285695843726, 0.00017480015346372276, 0.0001378277079923815, 7.633724223309769e-05, 
7.402155371667622e-12, 4.965827303819483e-11, 4.965437612207606e-11, 4.9650849634359917e-11, 4.9644769875527325e-11, 4.9642155753222594e-11, 4.963978972099507e-11, 4.963570972671434e-11, 4.9633955095898954e-11, 4.963236680374209e-11, 4.9786226003858936e-11, 5.025590754635561e-11, 5.2088158356424624e-11, 5.771888610298373e-11, 5.913705369744592e-11, 5.928091578082756e-11, 5.763067279392684e-11, 5.6569626886655146e-11, 5.565386489883837e-11, 5.4683841907138966e-11, 5.464633867216354e-11, 5.48062925637533e-11, 5.5366378165935806e-11, 5.5654855288171645e-11, 5.587381381126671e-11, 5.6044868611528313e-11, 5.602678592850865e-11, 5.5972769367126994e-11, 5.584042230149203e-11, 5.5791986607535523e-11, 5.575601417531867e-11]
bodono commented 1 week ago

Oh no! Can you dump the data down to csv again for one of them?

kkzhang3 commented 1 week ago

scs_err_csv.zip

bodono commented 1 week ago

Ok I have been able to reproduce this, as far as I can tell it is just numerical problems when the values are very small producing inconsistent results between different hardware. I will try and improve the robustness of the infeasibility checks by ensuring the things being compared are at some minimum value.

kkzhang3 commented 1 week ago

Oh, Thanks! Looking forward to the new version! I can provide more unexcepct infeasible examples. Let me know if it is necessary.

bodono commented 5 days ago

I believe I have fixed the issue in this PR: https://github.com/cvxgrp/scs/pull/280

I added the data files you included as tests, is this ok? If not I can remove them.

kkzhang3 commented 5 days ago

It's OK. I'll try the new version later.

kkzhang3 commented 3 days ago

Could you update a pip release?

bodono commented 1 day ago

Yes, will do as soon as I can figure out what's wrong with https://github.com/bodono/scs-python/pull/103