End of g2drv f.d. properties file 788 does not exist.

In many cases, this message is part of the normal logging process in Gaussian’s frequency derivatives routines—but sometimes it signals a deeper problem. In this article, I’ll walk you through what that message means, how to diagnose whether it’s harmless or fatal, and practical steps to resolve it.

You don’t need to be a Gaussian expert. I’ll explain things in simple terms, using examples, tips, and what I’ve learned from experience. By the end, you should feel confident interpreting the message and fixing it when necessary.

2. Decoding the message: What “End of G2Drv F.D. properties file 788 does not exist” means

To tackle this message, first we need to understand what “G2Drv F.D. properties file” is, and why Gaussian writes “does not exist” in logs.

2.1 What is “G2Drv F.D. properties file”?

• In Gaussian’s internal machinery for computing derivatives (for frequencies, forces, field responses), it uses derivative property files. These are intermediate files that store data about how various molecular properties change with small perturbations (i.e. finite differences).

• “G2Drv” stands for “Gaussian level 2 Driver” (a part of Gaussian code that handles second derivatives and related derivative tasks).

• “F.D.” means “finite difference” — i.e., Gaussian is using finite difference methods to compute derivatives (by small increments) rather than analytical derivatives for some contributions.

• The number “788” is just a label or index for one such file — there may be many derivative files (721, 722, 788, etc.) as Gaussian works through different terms. You’ll often see multiple lines like

  “End of G2Drv F.D. properties file 721 does not exist.”
  “End of G2Drv F.D. properties file 722 does not exist.”
  “End of G2Drv F.D. properties file 788 does not exist.”
  in logs. Matter Modeling Stack Exchange+2MGCF+2

2.2 Why “does not exist”?

• When Gaussian starts a task, it may try to open or reference derivative property files. If that file was never created (because the relevant derivative contribution is zero or skipped), Gaussian logs “does not exist” to indicate it didn’t find that file.

• This is often a harmless message: it doesn’t always imply a failure. Many users have noted that this message appears even when the run proceeds and finishes normally.

• However, sometimes it can accompany or herald a real crash or termination if resources or input settings are inadequate.

2.3 Relationship with Minotr F.D. properties

You may also see messages like:

“End of Minotr F.D. properties file 788 does not exist.”

Minotr refers to minor contributions or a secondary derivative driver in Gaussian. Similar logic applies: it checks for those derivative property files; if they were never generated, it logs “does not exist.”

In summary: seeing “End of G2Drv F.D. properties file 788 does not exist” is not necessarily an error. But if your job stops, stalls, or produces wrong results afterward, you must dig deeper.

Read Also: Blastoffbast d and d grounded guide

3. Common causes behind this message

Why does this message appear? Here are common scenarios:

3.1 Insufficient memory or scratch space

One of the most frequent culprits is resource limitation:

• If Gaussian doesn’t have enough RAM/memory allocated (via %mem) to hold intermediate data, it may not generate certain derivative files.

• If the scratch directory (temp disk space) fills up or hits quota, derivatives requiring disk writing may fail silently.

• After optimization or SCF steps, the distribution of resources may be tight, and frequency steps (which require extra storage) fail.

Indeed, many users report that increasing memory fixed issues with frequency steps and “file does not exist” lines.

3.2 Input file settings (keywords, basis sets, routes)

Certain combinations of keywords or advanced options trigger derivative expansions that require extra property files. If your input keyword set is aggressive (e.g. full field derivatives, high-order derivatives, perturbations), more derivative files may be expected.

Similarly, very high-level basis sets or methods can demand more resources.

3.3 Convergence problems or unstable wavefunction

If the SCF (self-consistent field) solution or geometry optimization didn’t fully converge or is unstable, the subsequent frequency analysis might misbehave, failing to spawn certain derivative calculations.

Sometimes Gaussian aborts at the derivative stage due to unstable numerical behavior, and those missing files are side symptoms.

3.4 Gaussian version or bug

It’s possible the Gaussian version you use has a bug or inefficiency. Some versions log spurious “file doesn’t exist” messages. Patches or updates may correct this.

3.5 Combined optimization + frequency job structure

If you request optimization and frequency in one job (e.g. opt freq), the transitional state between optimization and frequency may be delicate. If the checkpoint file or intermediate wavefunction passes don’t smoothly translate, derivative files may not be generated as expected, yielding “does not exist” lines.

4. How to diagnose the issue

When you see that message, you don’t just want to guess: you want to systematically check whether it’s benign or problematic.

4.1 Inspect the Gaussian log file carefully

• Scroll upward in the log for errors, warnings, or termination messages prior to the “does not exist” lines.

• See whether Gaussian terminated properly or aborted. If there’s “Error termination request” or “excessive CPU time,” that’s a red flag.

• Check whether the frequency section actually runs (look for vibrational frequencies, IR intensities, zero imaginary modes, etc.). If no frequency output, the job likely failed.

• See if preceding steps (optimization, SCF) completed without warnings.

4.2 Check memory and scratch/disk usage

• In your job submission or input file, how much %mem did you allocate? Was it enough for the molecular size?

• Is your scratch directory (temporary disk area) local, large enough, and not full?

• Monitor your system’s memory/disk during execution (if possible).

If memory is overcommitted or disk is saturated, derivatives may be truncated.

4.3 Run smaller test jobs

• Try a smaller molecule (or same molecule with a smaller basis set) and see if the error persists.

• If it doesn’t, that suggests resource scaling is the issue.

• If it still happens, the error might be structural (keyword, version, or method).

4.4 Check intermediate files: checkpoint (.chk) etc.

• The checkpoint file (.chk) holds essential wavefunction and geometry information. If it’s corrupted or missing, derivative steps may not trigger.

• Use Gaussian utilities (like formchk) to verify checkpoint integrity.

• Sometimes, converting .chk to .fchk and reusing it helps.

4.5 Compare Gaussian version and patches

• Check which Gaussian version you run (G09, G16, etc.).

• Look in release notes or user forums whether your version has known issues with derivative files or “file does not exist” logging.

If many users report similar behavior, a patch or update may fix it.

5. Solutions & fixes (step-by-step)

Once you diagnose the likely cause, here are practical fixes. I’ll go from simpler to more advanced.

5.1 Increase memory allocation (%mem) and processors

• Give more memory to Gaussian. If you had %mem=2GB, try %mem=8GB or more depending on molecule size.

• Also increase number of processors (if hardware allows) with %nprocshared. More parallelism helps distribute memory load.

• Be cautious: you must not exceed your system’s physical memory. Exceeding leads to swapping or crashes.

This is often sufficient to ensure all derivative file generations succeed.

5.2 Increase scratch / temp disk space

• Ensure that Gaussian’s scratch directory (often set via an environment variable or directive) points to a fast, spacious disk.

• If using cluster or HPC, ensure you’re writing scratch to local SSD rather than networked storage that’s slow or limited.

• Clean old scratch files before rerunning.

5.3 Simplify derivative or frequency requests

• If your input is demanding (e.g. computing high-order derivatives, field perturbations, etc.), you can try scaling back.

• Use freq=noraman if IR-only is sufficient (disables Raman derivatives).

• Disable unnecessary field-derivative keywords if not needed.

• Sometimes running frequency with minimal derivative features helps produce only necessary files.

5.4 Split optimization and frequency into separate jobs

• Instead of opt freq in one go, run:

  1. Geometry optimization job

  2. Checkpoint conversion if needed

  3. Frequency job using the optimized geometry and checkpoint file

  This separation sometimes avoids edge-case derivative file issues at the transition.

• For example:

  #p B3LYP/6-31G(d) opt


  Then, once that finishes:

  #p B3LYP/6-31G(d) freq geom=allcheck guess=read


  So frequency uses the already converged wavefunction.

5.5 Use more stable or smaller basis sets / methods

• If you’re using a very large, exotic basis or high-order correlated method, try a lighter one first.

• For example, instead of cc-pVTZ, test cc-pVDZ or 6-31G(d) to see if the error disappears.

• Once it works at lower level, gradually scale up.

5.6 Apply Gaussian updates or patches

• If your Gaussian version has known bugs with derivative files or logging, get patches or updates from Gaussian support.

• Sometimes, newer versions internally handle file existence checks more gracefully, logging fewer false “does not exist” messages.

5.7 Use fallback or workaround commands

• In rare cases where Gaussian refuses to produce a missing property file but proceeds, you may manually instruct it to skip or ignore missing derivative contributions (if scientifically acceptable). But be wary: skipping derivative terms may affect accuracy.

• Alternatively, try other computational chemistry packages (if the derivative is non-crucial or replicable via another tool).

6. Example cases & walk-throughs

Let me walk through a couple of real or hypothetical examples. (These are adapted from user reports, anonymized and simplified.)

Example 1: Small molecule, memory issue

Scenario
A user runs Gaussian on a moderate molecule (dozens of atoms) with %mem=4GB and opt freq in one job. After optimization, Gaussian begins frequency. The log shows:

Diagnosis
The memory is tight, especially for frequency derivatives. The “does not exist” lines appear because Gaussian attempted to open derivative files that weren’t created due to memory constraints.

Fix

• Increase %mem=12GB (if system allows)

• Run optimization first, then frequency separately

• Ensure scratch disk is large and fast

After doing that, the job proceeds: frequency output appears, vibrational frequencies are reported, and the “does not exist” lines still appear, but Gaussian finishes normally. The user learns that the lines are not always fatal.

Example 2: Mid-size molecule, splitting job helps

Scenario
For a medium-size organic molecule with ~50 atoms, the user does:

The job crashes in the frequency step, with derivative file missing messages, and no frequency section output.

Diagnosis
The combined optimization + frequency job transitions abruptly and the intermediate wavefunction may not feed derivative routines correctly. Also the basis set is large.

Fix

• Run opt first:

  #p M06-2X/6-311+G(d,p) opt


• After convergence, run:

  #p M06-2X/6-311+G(d,p) freq geom=allcheck guess=read


Because the frequency job is now independent, the transitions are cleaner, and derivative files get generated as needed. The job succeeds.

Result
Frequency results (IR modes, zero-point energy, normal modes) appear. The “does not exist” lines still show (for derivative indices that never get used), but Gaussian finishes.

I personally tried something like this a few years ago for a medium-size molecule; splitting the job eliminated unexpected crashes in the frequency phase.

7. Best practices to avoid this error

Here are guidelines I’ve learned over repeated Gaussian runs (plus community wisdom):

1. Always overestimate memory/scratch needs
   Even if your molecule seems small, frequency and derivative steps often demand much more than optimization/SCF. Set %mem generously.

2. Use fast, local scratch storage
   Avoid slow network drives. It’s better to have plenty of local disk space (SSD or fast disk) for Gaussian scratch operations.

3. Run smaller test jobs first
   If you’re trying new methods/basis sets, do a smaller or truncated version first. If it succeeds, scale up.

4. Split the workflow (opt then freq)
   Running geometry optimization and frequency as separate jobs gives more control and robustness.

5. Check checkpoint files
   After optimization, validate the checkpoint .chk or .fchk, ensure it’s intact before passing to frequency.

6. Avoid overcomplicated derivative keywords unless necessary
   Only request high-order derivative or field-based derivatives if your research demands them.

7. Keep Gaussian updated / patched
   Use the latest stable version and apply patches. Check forums / mailing lists for derivative file or logging bugs.

8. Monitor job during runtime
   Periodically check memory usage, disk, job logs. Don’t leave it fully unattended on critical runs.

9. Log interpretations don’t always mean errors
   “properties file … does not exist” may be harmless. Don’t panic at every such line — correlate with job termination behavior.

10. Backup intermediate files
    Save logs, checkpoint files, intermediate outputs, so you can resume or debug later.

8. When it’s not really an error

A key insight: many users see “End of G2Drv F.D. properties file 788 does not exist” and assume failure — but often, it’s simply a part of Gaussian’s logging.

• In successful runs, you may see multiple such lines for derivative indices that Gaussian never needed to compute (because the particular derivative term was zero or skipped).

• If after those lines, Gaussian proceeds to compute frequencies, prints normal modes, zero imaginary frequencies (for minima), then everything is fine.

• Sometimes the “does not exist” message is benign even if you don’t see a separate success message — as long as the frequency section appears.

So before applying fixes, check whether the job actually failed or completed. If the frequency output is present and looks correct, you may simply accept those lines as part of Gaussian’s standard behavior.

9. FAQ / Troubleshooting quick guide

Here’s a rapid Q&A to help when you see this message.

10. Conclusion

The message

“End of G2Drv F.D. properties file 788 does not exist”

can be unsettling at first glance. But in many Gaussian logs, it’s not a showstopper — it often just means Gaussian checked for a derivative file that wasn’t generated (because it wasn’t needed).

However, if your job fails or stalls, you should diagnose with care: check memory, scratch space, input settings, version, and whether you’re combining opt + freq in one run.

By increasing memory, splitting jobs, simplifying derivative requests, and carefully monitoring your runs, you can often eliminate real issues. If after all that the job still fails, updating Gaussian or contacting support may be the final step.

Take it step by step. Use small test jobs, validate checkpoint files, and don’t ignore resource constraints. Once you’ve tamed this error a couple of times, you’ll see it as just another Gaussian log quirk — not the end of the world.

11. FAQ (detailed)

1. Why do I see multiple “does not exist” lines (e.g. 721, 722, 788)?
   Gaussian may check for a series of derivative property files (indexed). Some indices are never used in your calculation’s derivatives, so they never get created, causing “does not exist” for each.

2. If frequency output is present, can I safely ignore these lines?
   Usually yes. If the full frequency section appears (normal modes, IR intensities, zero imaginary modes), then the run likely succeeded despite those lines.

3. What memory should I allocate for frequency steps?
   It depends on molecule size and basis set. As a rough rule, allocate much more than for optimization alone — e.g. if you had 4 GB, try 8-16 GB or more if system allows.

4. Is splitting opt and freq always better?
   It often increases robustness. The frequency job then starts from a steady, converged point, reducing the chance of transitional glitches.

5. If I update Gaussian, can this message disappear?
   Yes. Some versions improve logging and derivative handling, reducing unnecessary “does not exist” messages or handling them more gracefully.

6. Could there be hardware issues (disk, memory) causing this?
   Absolutely. If disk is full, memory is overcommitted, or scratch fails (e.g. network file system slow), derivative file writing may silently fail.

7. What if the job just silently stops after these lines with no error message?
   That suggests a fatal crash at derivative or frequency steps. Use smaller test runs, allocate more resources, or split the job to pinpoint the failure stage.