Hello everyone and @itascan @msepu @itascaYang
I am currently facing an issue with my model. The model converges when I use model solve 1e-2. I have defined several history points to monitor parameters at different locations. However, when I try to plot the history charts, no curves appear.
Additionally, when I export the results to a CSV file, it contains only a single value corresponding to one step.
To investigate further, I ran the model using model cycle 5000. In this case, the history curves appear, and the exported file contains values over multiple steps. However, the results differ significantly from my theoretical calculations.
This leads me to believe that 5000 cycles may not be sufficient for the model to reach equilibrium. I would like to ask if there are any guidelines or criteria to determine the appropriate number of cycles required for a model to reach equilibrium.
Any suggestions or insights would be greatly appreciated.
Thank you for your time and support.
While 1e-2 might sound like a small number, in the context of the unbalanced force ratio; 1e-3 is generally considered a preliminary convergence criterion and for most engineering applications, a ratio on the order of 1e-5 is the standard for equilibrium (although some models may require something even smaller than this).
Also note that the default history increment is every 10 steps (this can be changed with history interval <i_step> command). It may be that your initial model hasn’t reached 20 steps. You can see number of steps in the console after a solve or cycle or step command. This can also be displayed in any plot by enabling step in the plot legend settings, or by checking the object tree.
To ensure you have reached equilibrium and minimize solve times, instead of relying solely on a standard number, it’s always good practice to check these three “safety signs”:
-
The History Plot: Plot the “Max Unbalanced Force Ratio” vs. “Step.” The curve should be flat (asymptotic). Note that if this curve flattens out at a value above your target threshold then the model may have reached a state of continuous plastic flow: your slope is sliding, your tunnel is collapsing, or your foundation has failed.
Velocity: Check the maximum velocity in the model. It should be approaching zero (e.g., 10^-6 or less).
-
Displacement: Monitor a “target” node in the zone of interest. If the displacement is still climbing linearly when the model stops, you haven’t reached equilibrium. If it becomes stable for a long period of time, you can reduce the target to solve faster.
Here’s a summary of things to consider:
Is the default maximum unbalanced force ratio for model solving 1e-5? When the maximum unbalanced force ratio is 1e-2, the model may reach the “requirement” after only a very few steps. Additionally, based on my experience in FLAC, the plastic zones stabilize when the maximum unbalanced force ratio is below 1e-3. In my simulations, if I don’t want to spend too much time on in-situ stress balance, I usually set the maximum unbalanced force ratio to 1e-4.
Hello @ddegagne
Thank you for your reply.
Basically, the issue I am facing is related to solving my model. Initially, I used a convergence criterion of 1e-4, but the model kept running continuously for 2–3 hours (even longer) without achieving convergence. After that, I tried 1e-3, but the model still did not converge. Finally, when I used 1e-2, the model completed the run successfully.
However, another issue appeared in this case. I used the history command to monitor some parameters. After the model run was completed, I attempted to plot the results using the history chart, but no curve was displayed. When I exported the data as a CSV file from the empty plot, it showed only a single step of data, and in some cases, no data was recorded at all.
At this point, I am quite confused about how to resolve this issue.
Hello @itascaYang
Thank you for your reply.
Basically, the issue I am facing is related to solving my model. Initially, I used a convergence criterion of 1e-4, but the model kept running continuously for 2–3 hours (even longer) without achieving convergence. After that, I tried 1e-3, but the model still did not converge. Finally, when I used 1e-2, the model completed the run successfully.
However, another issue appeared in this case. I used the history command to monitor some parameters. After the model run was completed, I attempted to plot the results using the history chart, but no curve was displayed. When I exported the data as a CSV file from the empty plot, it showed only a single step of data, and in some cases, no data was recorded at all.
At this point, I am quite confused about how to resolve this issue.
Hi @Sudipta,
When your Max Unbalanced Force Ratio vs. Step plot plateaus (flattens out) like that at a value above your target threshold, it could mean:
- The model has reached a state of continuous plastic flow (i.e., your slope is sliding, your tunnel is collapsing, or your foundation has failed). The “unbalanced force” remains constant because the model is physically in motion.
- Numerical “noise” or precision limits. If the plateau is very low (e.g., 1x10^-6) but won’t hit 1x10^-7, you might simply be hitting the limits of double-precision math or the “noise” of the calculation. If it plateaus at such a low level, you can likely stop the solve.
- Poor grid quality or stiffness contrast. If you have “bad” zones (very high aspect ratios or inverted faces) or a massive difference in stiffness (e.g., a steel bolt next to soft clay), this can create “local” imbalances that stays stuck, preventing the “global” ratio from dropping further. Also, if attaching grids, one or more nodes may not have attached (tolerance issue).
A couple of questions:
- Which code (FLAC, FLAC2D, or FLAC3D) and which version are you running?
- Can you please explain what you mean by “1×10−41×10−4”? Do you mean “1x10-4” or “1E-04”?
- How many steps did each solve ratio go before you stopped model runs?
A couple suggestions:
- Make sure you are using the latest version of the software. It’s possible there’s a bug somewhere that has already been fixed. Review the update logs on the software download page.
- Double check your material (zones, SELs, etc.) properties - check for typos. Review data file and/or plot properties for visualize feedback. Confirm units (using consistent set of units, not mixing metric and imperial units - gravity at 9.81 not 32 or vice versa).
- Double check boundary conditions (plot gp fixities) and stress or force signs (+ve vs. -ve).
- Plot displacements and look for displacements in odd locations (along boundaries, at a gridpoint away from excavations.
- Plot Zone Local Force Ratio to visualize local stability (good way to ensure that every part of your model is stable, not just the “average.”)
- Plot Zone Convergence to visualize whether the gridpoint forces and the stresses are still changing. Useful in complex multi-physics or high-plasticity models where the force ratio might “flutter” but the model is technically stable.
- Try reducing history interval to 1 for your solve runs (not step runs).
You may also want to review “Reaching Equilibrium” in the documentation: Reaching Equilibrium — Itasca Software 9.6 documentation.
Cheers - Dave
If your parameter settings are unreasonable, or if you conduct large-scale excavation (with an excessively large range covered by null elements), it may lead to non-convergence.
