<div dir="ltr"><span style="font-size:12.8px">This makes no sense: -linear_snes_rtol 1.e-60</span><br><div><span style="font-size:12.8px"><br></span></div><div><span style="font-size:12.8px">This is the relative tolerance and so 1.e-14 with double precision about as low as you want to go. Right? I'm not sure why this converges at all.</span></div><div><span style="font-size:12.8px"><br></span></div><div><span style="font-size:12.8px">Anyway, It looks like your problem get hard after the first time step. Run with one processor and 'lu' instead of 'asm' for the pc_type. This is a direct solver. Then I would look at your solution and make sure that it looks OK. Often poor iterative solver performance is correlated with a simulations that is bad.</span></div><div><span style="font-size:12.8px"><br></span></div><div><span style="font-size:12.8px">And what kind of equation and discretization is this?</span></div><div><span style="font-size:12.8px"><br></span></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Mon, Jul 17, 2017 at 10:16 AM, 李瑞 <span dir="ltr"><<a href="mailto:lirui319@hnu.edu.cn" target="_blank">lirui319@hnu.edu.cn</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><br>
Dear professor or engineer:<br>
I recently attempt to solve the system of equation that consist of two sub equations,which are respetivly spatially and timely dependant.The structure of equations are likely : F_1(x,y)=0;F_2(w(t),x)=0.and y can be explicitly obtained in each time step.We need to pay attention to the first equation is a linear equation and the second a nonlinear equation dependent on time.Getting W(t0) by sequentially obtain y and x.<br>
When I solve the equation F1 in first time step,the result was convergent,but in the second step was not.The answer reveals that:<br>
>>>In the first time step:<br>
....<br>
126 KSP Residual norm 6.505554772167e-03<br>
127 KSP Residual norm 2.563866095029e-03<br>
128 KSP Residual norm 5.365504095451e-04<br>
129 KSP Residual norm 9.699344087450e-05<br>
130 KSP Residual norm 2.793539286585e-05<br>
131 KSP Residual norm 8.413249813721e-06<br>
132 KSP Residual norm 2.488394296066e-06<br>
133 KSP Residual norm 1.055822777756e-06<br>
134 KSP Residual norm 7.820629842889e-07<br>
Nonlinear solve converged due to CONVERGED_ITS<br>
<br>
>>>In the second time step :<br>
....<br>
294 KSP Residual norm 2.632117164436e+03<br>
295 KSP Residual norm 2.627315190869e+03<br>
296 KSP Residual norm 2.626944159914e+03<br>
297 KSP Residual norm 2.625073489387e+03<br>
298 KSP Residual norm 2.624035038684e+03<br>
299 KSP Residual norm 2.623956156495e+03<br>
300 KSP Residual norm 2.623696336837e+03<br>
Nonlinear solve did not converge due to DIVERGED_LINEAR_SOLVE<br>
<br>
The following code is typing my makefile.<br>
-linear_ksp_monitor -linear_snes_converged_reason -linear_snes_atol 1.e-4 -linear_snes_rtol 1.e-60 -linear_ksp_type gmres \<br>
-linear_ksp_max_it 300 -linear_ksp_atol 1.e-6 -linear_ksp_gmres_restart 90 -linear_ksp_rtol 1.e-60 \<br>
-linear_ksp_pc_side right -linear_pc_type asm -linear_pc_asm_type restrict -linear_pc_asm_overlap 1 \<br>
-linear_sub_ksp_type preonly -linear_sub_pc_type lu \<br>
<br>
Is there anything wrong,Pc_type or Krylov method?Can you give me some detailed suggestions or other ways to solve my obsession?Thank for your heartful help!<br>
</blockquote></div><br></div>