Leginon

First, regarding acquiring Bright Image for K2 counting mode, your procedure is correct. The only part I think need clarification is the dose rate used in this procedure. The 100 second accumulated exposure time was based on our experience with 8 electrons per pixel per second dose rate. If you have used a lower dose rate, you should increase the total exposure time so you get good statistics.

The way to tell if you've gotten a statistics sound bright image is to compare it with the one you get when you acquire gain reference in Digital Micrograph, and or this one:
http://emg.nysbc.org/attachments/2352
The grow zone (global gradient) can only be visualized if you have accumulate enough counts. Every camera sensor should have something like it even if not in the same direction.

Second regarding acquiring image series for doing goniometer model, you are quite right that you should reduce the magnification to where K2 camera uses. 550x appears to be too high. I see that you have to break up the measurement, probably because a grid bar darkened the whole imaging area.

The interval number is not the cause of the bad calibration. 5e-06 m is close to optimal for picking up the necessary the high harmonic terms of the oscillation we have known from our older FEI Tecnai scopes. You can vary from 3-6 um and still get reasonable results. Your graph is hard to read because each half of your measurement covered short of two complete cycles of the expected model base harmonic term ( ~ 60 um). Even though the program can pick up the oscillation fine even with this, human eyes won't be able to do the same. To improve this, find an area of the grid that can give you longer uninterrupted measurement, or, lower the magnification so that the grid bar does not block the view, or rotate or use a grid has more opening area.

However, the graph does show overfitting. New generation of FEI microscopes no longer has the oscillation behavior, probably since around 2006. You should check your old graph for the same instrument and goniometer axis to see this applies to your instrument. The y-axis in the graph is the deviation from linearity normalized so that if the movement is linearly responding to the instruction sent by the program, the y-value would remain at 1.00. 0.99 means that the resulting movement is 1 % short of what it is asked to moved to. Looking at the plot you attached, the deviation is mostly within 1 % (meaning that if you ask it to move 5 um, it moved to 4.95 um). It would be unnecessary to fit this with an oscillating function. The old goniometers had 8-20% deviation at places. In your current case, 0 term fit as is done normally done for y-axis is sufficient. You can even totally use the plain "Stage Position" matrix calibration since 0 order harmonic function in both axis is no different from linear fits in each direction.

Hope these answer all of your questions.

Hello Anchi,

Thank you very much for your reply.

For the Bright/Dark reference images, I did see the global gradient on the bright reference images. And the corrected image looks very homogenous.

For the GonioModeling, I tried decreasing the magnification from 330X to 110X and now it seems that the program could recognize the image features. And I have attached the results as the "goniometergraph_x_mag110X" and "goniometergraph_y_mag110X". The results still display a random oscillation around 1.000. Since all these points were very close to 1, I assumed that the GonioModeling went well and finished the other parts for other magnification.

However, after I set the acquisition "tomo" target in "hl" during my data collection today, the Leginon program could not find the correct target position and display an alarm message "Move Failed. skipping acquisition at this target"

And I also checked "Leginon\Navigation" and the Navigation window also shows a full lists of move errors.
I also attached the Nagivation_error as the attached file.

Would you mind checking the attached files and help us about this movement issue?

Thanks again,
Long

Long,

With the deviation from linearity so small, I recommend that you refit the model with 0 term. It reduces over fitting error. Think about the scattering you see as normal distributed sampling of the linear response. The more you try to fit according to the fluctuation, the worse it would be since it will now producting fluctuation of values according to the curve which would be overall wider than the straight line.

As for the navigation error, Please take a screenshot of the correlation map. Have you used this microscope for Leginon Tomography before? Do you know if this is worse than before? In other words, I want to rule out compustage problem. Looking at the values change of the last step, it appears that the stage overshot. Try some reproducibility test on the compustage, moving between two saved coordinates to see if it can reproduce it o.k.

If you think that the correlation peak is not correctly located, you can increase the exposure time a bit. If it still does not look right, you will need to send me two mrc images that I can test the correlation with.

Hi Anchi,

I would like to know what is your opinion of the attached measurements of Modeled Stage Calibration for our Krios/GIF/K2 system. Following your discussion with Long, I assume that the best fit for the is a 0 order function. Am I right?
Thanks

Mario

Yup!