Error while running - NaN

Errors
1

Hello everyone,

I’m trying to run a simulation with PyFR, and I’m gettig error for NaN as the runtime is geting to the delta time which was specified.

Here’s the ini file:

[backend]
precision = single
rank-allocator = linear

[backend-cuda]
device-id = local-rank
;devide-id = round-robin
mpi-type = cuda-aware

[constants]
ac-zeta=50
;ac-zeta = 0.4
gamma = 1.4
Uin = 4.6
Vin = 0
Win = 0
Pc = 101325
Pr = 0.72
nu = 0.000015

[solver]
system = ac-navier-stokes
order = 3
anti-alias = flux, surf-flux

[solver-time-integrator]
formulation = dual
scheme = backward-euler
pseudo-scheme = rk45
controller = none
tstart = 0.0
tend = 24
dt = 5e-5
;dt = 0.0016
atol = 0.00001
rtol = 0.00001
pseudo-dt = 1e-6
;pseudo-dt = 4e-5
pseudo-niters-max = 20
pseudo-niters-min = 3
pseudo-resid-tol = 1e-4
pseudo-resid-tol-p = 1e-4
pseudo-resid-norm = l2
pseudo-controller = local-pi
safety-fact = 0.8
min-fact = 0.98
max-fact = 1.01
pseudo-dt-max-mult = 2.5

[solver-interfaces]
riemann-solver = rusanov
ldg-beta = 0.5
ldg-tau = 0.1

[solver-dual-time-integrator-multip]
pseudo-dt-fact = 1.75
cycle = [(3, 1), (2, 1), (1, 1), (0, 3), (1, 1), (2, 1), (3, 3)]

[solver-interfaces-line]
flux-pts = gauss-legendre
quad-deg = 9
quad-pts = gauss-legendre
[solver-interfaces-line-mg-p2]
flux-pts = gauss-legendre
quad-deg = 7
quad-pts = gauss-legendre
[solver-interfaces-line-mg-p1]
flux-pts = gauss-legendre
quad-deg = 5
quad-pts = gauss-legendre
[solver-interfaces-line-mg-p0]
flux-pts = gauss-legendre
quad-deg = 3
quad-pts = gauss-legendre

[solver-elements-tri]
soln-pts = williams-shunn
quad-deg = 9
quad-pts = williams-shunn
[solver-elements-tri-mg-p2]
soln-pts = williams-shunn
quad-deg = 7
quad-pts = williams-shunn
[solver-elements-tri-mg-p1]
soln-pts = williams-shunn
quad-deg = 5
quad-pts = williams-shunn
[solver-elements-tri-mg-p0]
soln-pts = williams-shunn
quad-deg = 3
quad-pts = williams-shunn

[solver-elements-quad]
soln-pts = gauss-legendre
quad-deg = 9
quad-pts = gauss-legendre
[solver-elements-quad-mg-p2]
soln-pts = gauss-legendre
quad-deg = 7
quad-pts = gauss-legendre
[solver-elements-quad-mg-p1]
soln-pts = gauss-legendre
quad-deg = 5
quad-pts = gauss-legendre
[solver-elements-quad-mg-p0]
soln-pts = gauss-legendre
quad-deg = 3
quad-pts = gauss-legendre

[solver-elements-hex]
soln-pts = gauss-legendre
quad-deg = 9
quad-pts = gauss-legendre
[solver-elements-hex-mg-p2]
soln-pts = gauss-legendre
quad-deg = 7
quad-pts = gauss-legendre
[solver-elements-hex-mg-p1]
soln-pts = gauss-legendre
quad-deg = 5
quad-pts = gauss-legendre
[solver-elements-hex-mg-p0]
soln-pts = gauss-legendre
quad-deg = 3
quad-pts = gauss-legendre

[solver-elements-pri]
soln-pts = williams-shunn~gauss-legendre
quad-deg = 9
quad-pts = williams-shunn~gauss-legendre
[solver-elements-pri-mg-p2]
soln-pts = williams-shunn~gauss-legendre
quad-deg = 7
quad-pts = williams-shunn~gauss-legendre
[solver-elements-pri-mg-p1]
soln-pts = williams-shunn~gauss-legendre
quad-deg = 5
quad-pts = williams-shunn~gauss-legendre
[solver-elements-pri-mg-p0]
soln-pts = williams-shunn~gauss-legendre
quad-deg = 3
quad-pts = williams-shunn~gauss-legendre

[solver-interfaces-tri]
flux-pts = williams-shunn
quad-deg = 9
quad-pts = williams-shunn
[solver-interfaces-tri-mg-p2]
flux-pts = williams-shunn
quad-deg = 7
quad-pts = williams-shunn
[solver-interfaces-tri-mg-p1]
flux-pts = williams-shunn
quad-deg = 5
quad-pts = williams-shunn
[solver-interfaces-tri-mg-p0]
flux-pts = williams-shunn
quad-deg = 3
quad-pts = williams-shunn

[solver-interfaces-quad]
flux-pts = gauss-legendre
quad-deg = 9
quad-pts = gauss-legendre
[solver-interfaces-quad-mg-p2]
flux-pts = gauss-legendre
quad-deg = 7
quad-pts = gauss-legendre
[solver-interfaces-quad-mg-p1]
flux-pts = gauss-legendre
quad-deg = 5
quad-pts = gauss-legendre
[solver-interfaces-quad-mg-p0]
flux-pts = gauss-legendre
quad-deg = 3
quad-pts = gauss-legendre

[soln-plugin-pseudostats]
flushsteps = 10
file = pseudostats.csv
header = true

[soln-plugin-nancheck]
nsteps = 1

[soln-plugin-writer]
tstart = 12.0
dt-out = 0.05
basedir = /home/amir/
#basedir = .
basename = Urban2-{t:.4f}

[soln-ics]
u = 4.6
v = 0.0
w = 0.0
p = Pc

[soln-bcs-walls]
type = no-slp-wall
u = 0
v = 0
w = 0

[soln-bcs-inlet]
type = ac-char-riem-inv
ac-zeta = 50
;ac-zeta = 5
rho=1.185
p = Pc
u = Uin
v = Vin
w = Win

[soln-bcs-outlet]
type = ac-char-riem-inv
ac-zeta=50
;ac-zeta = 5
rho=1.185
p = Pc
u = Uin
v = Vin
w = Win

[soln-bcs-farfield]
type = ac-char-riem-inv
ac-zeta=50
;ac-zeta = 5
p = Pc
u = Uin
v = Vin
w = Win

Please advice how to solve the problem.

2

In the majority of cases this is caused by:

  1. An incorrect start-up procedure (starting a simulation directly at a high-order rather than gradually increasing it).
  2. A mesh which is too coarse.
  3. A time step which is too larger.

Without knowing more about your simulation it is difficult to say what is going on.

Regards, Freddie.

3

Hi Freddie,

how do you suggest to rewrite the ini file for gradually increased startup?

Thanks.

4

PyFR allows you to restart a simulation with a different ini file. So you start with an initial file say, at order = 1. Then, run the simulation for some period of time, then restart from the final solution with a new ini file with order = 2. This will typically be more robust.

Regards, Freddie.

5

Thanks Freddie.

I tried your method and it was running ok on p=1 & p=2 with restarting. However, when I restarted the simulation with p=3 , there was no convergence i.e. NaN.
To be honest, I’ve conducted the simulation first with pressure P = 1.013 by mistake and the simulation worked properly, so I don’t think there is any coarse meshing issues.
I don’t know how to proceed on this, maybe you can help me.

Best regards,
Amir

6

How does the flow look at p = 2? Is it well developed? When you restart at p = 3 how does the flow look before it NaNs? Usually, this will give you an idea of where you are under-resolved.

Regards, Freddie.

7

I apologize, the simulation didn’t restarted at p=2, not at p=3. However, I ran the simulation with p = 1 for quite long time, so it should converege.

Also, I tried to restart from the simulation with BC P = 1.013 (where it worked propely with p = 3) and replace the pressure value to 1.013e5, but it also didn’t convereged. Does changing the pressure value is vital for convergence?

8

If you want to make any progress you need to actually start looking at the flow. As I mentioned before the issue is almost certainly with your mesh; the fact a mesh works at a low polynomial order but not a higher one does not in itself mean much. Lower polynomial orders have more dissipation which can allow under-resolved simulations to remain stable. This goes away as the order is increase.

Arbitrarily changing the pressure is also unlikely to result in a meaningful outcome. The pressure should be fixed by the physics you are trying to replicate and your chosen set of units.

Regards, Freddie.