Maximum Kernel is Near Zero

[AbolfazlKhanMo]

Hi there,

I am using TomoATT to do some inversion. Before starting the inversion, I tested several number of grid points and did a small performance study given my dataset.

When it comes to inversion, for some reason, I get the following:

mode update starting ...
Error: max_kernel is near zero (less than 10^-12), check data residual and whether no data is used

[JingChen-Thu]

Hello, @AbolfazlKhanMo

Thanks for using TomoATT. This error indicates the sensitivity kernel is near 0. You can check the input parameter YAML file (usually "input_params.yaml")

The following parameter in Lines 184 should be set True. This option allows for the inversion of absolute traveltime.

This is the most possible reason for your error. If this problem is not addressed, you can send me the YAML file for further debugging.

By the way, did you successfully run the examples in TomoATT? You can also use the parameter file in the example by modifying a few necessary parameters, such as "domain", "source", "model", and parameters associated with "inv_grid".

(You can @JingChen-Thu (me) for rapid feedback here)

Sincerely,

Jing CHEN

[bru-melo]

@JingChen-Thu I am unfortunately struggling with the same error. I have been running successful inversions, now when I try to add new events to the same catalogue I have been using, and using the same input parameters I get the error:

mode update starting ...
Error: max_kernel is near zero (less than 10^-12), check data residual and whether no data is used
Error: max_kernel is near zero (less than 10^-12), check data residual and whether no data is used

I'm attaching the input yaml file.

Many many thanks in advance,

Bruna.
input_params_inv.yaml

[JingChen-Thu]

@bru-melo

Dear Bruna,

First, I am glad to know that you have successfully run the inversion. This means that your setup is somewhat correct, at least before adding new data.

I have checked your input yaml file, and it looks fine to me. I built an initial model based on the Crust1.0 model and your yaml file, then created a src_rec_file consisting only one absolute traveltime data, and ran a test inversion. The inversion works well, which also confirms that your yaml file is likely correct.

Here are several possible causes for the issue you encountered:

1. Sources or Receivers are located on the boundary
   I observed one case which may reproduce the same error: if a receiver (a station if swap_src_rec is set false, or an earthquake if swap_src_rec is set true) lies exactly on the boundary of the study region. In this situation, this receiver is regarded valid and no "out of region" error is triggered. However, the adjoint source is located on the boundary, and will be forced to 0, leading to zero value of the kernel.

However, this would lead to a zero kernel only if all sources and receivers are on the boundary, which is not common. Even though, I suggest that you check carefully whether the newly added events are exactly on or near the boundary of the study region (especially their depths).

2. Using synthetic data instead of observed data as input
   When we use TomoATT to do forward modeling, relocation, or tomography, synthetic traveltime file are generated in the output folder. For example, after relocation using TomoATT, we can obtain, e.g., src_rec_file_reloc_0200_obs.dat and src_rec_file_reloc_0200.data. The former contains observed data with updated locations, while the latter contains the synthetic data with updated locations. If src_rec_file_reloc_0200.data is mistakenly used as the input, the objective function value and kernel will be 0, leading to the same error as you reported.

In this case, please check the src_rec_file contains really observations rather than synthetic data.

Cased on your current description, I can only provide the above two guesses. I hope these comments can address your issue. If not, perhaps I need more information to diagnose the issue. It would be helpful if you can attach the complete output log file (the text log consisting of the ERROR).

Alternatively, you may send me the three input files to replicate the error:
src_rec_file
model.h5
yaml (you have attached).
For data confidentiality, you can send only a small subset that reproduces the error, to me via e-mail (jing.chen@ntu.edu.sg). Once I can replicate the error, I am confident that I can address it.

Finally, thank you very much for using TomoATT. If you have further problem, please feel free to @ me here.

Sincerely,

Jing CHEN

[JingChen-Thu]

@AbolfazlKhanMo @bru-melo Thank you very much for reporting this bug, and helping us address it. Now this bug has been fixed in this update #63

Please wait a little bit until the update is merged to the main branch. Once it is merged, you can reinstall TomoATT to apply the fix. We apologize for any inconvenience this have caused.

If you encounter any further issues, please feel free to @ me here.