25

After my PhD, I continued with a postdoc in my PhD lab. The topic and the numerical method was quite different than my PhD and I had to read up quite a bit of literature to start the new work. However, as I was also involved in other projects as well, I took around nine months to complete the work. While writing the paper, I have discovered that I have done a major mistake in my simulation parameters and model. The work cannot be published unless I correct the mistakes and run the simulations again which would take at least a month more.

I was asked to submit the manuscript for review by the advisor and other collaborators by next week. However, now I can't write the paper with the errors. And I have to rerun the entire work again.

What do I do now? I am almost a year into my postdoc, and I shouldn't have made such errors. How do I approach this and what do I tell my advisor and collaborators?

6
  • 11
    The situation obviously isn't ideal, but there doesn't seem to be much we can help with. There aren't many ways to say "I found a mistake and I need more time to redo some of the work", and it seems like you already know you need to let the other people know and fix the mistake. Based on your title, I assumed the error you found invalidated the entire year of work (or your PhD). Simply having to take a bit longer doesn't seem like a huge deal (unless you'd miss a once-a-year deadline or something, but you still wouldn't really have options). Also note: your advisor is there to advise you.
    – NotThatGuy
    Jun 30, 2021 at 8:27
  • 6
    Bad news doesn't get better with age, so don't delay in letting people know what's going on. But, speaking from experience, this kind of news is not as bad as you think (you found the error even before your collaborators even saw the paper!). In a few months, people will not remember this.
    – Andrew
    Jul 1, 2021 at 4:36
  • 3
    If the mistake you found was a natural mistake, and the finding of the mistake reflects deeper knowledge gained over the course of the year, perhaps that part of the research could be salvaged. Good papers have been written dissecting bad models. Jul 1, 2021 at 12:41
  • 1
    If you find a mistake now, that's a lot better than finding it later after everything has been published, as then you would have to go through some rigmarole of withdrawing the article and so on.
    – Tom
    Jul 1, 2021 at 15:09
  • 2
    could you throw more computing power at running your simulations? Might not be cheap but this is an ideal use case for cloud computing (assuming your simulations are parallelisable). Jul 2, 2021 at 10:29

5 Answers 5

89

It's always best to find errors as early as possible. You've found this one before rather than after publication, so that's a big win. There's nothing magical about any step in one's academic career that makes them immune to making errors: everyone makes mistakes, no matter how many degrees they have or years of experience or papers they've published. It's a fair bet that the most senior researchers have made the most errors, simply by having had more time to accrue them.

Let your advisor and collaborators know what happened, and start running the corrected simulations as soon as possible. If you can update them with a timeline of when it would be feasible to submit the new results, then do so.

They may be disappointed by the delay, but they'll prefer this rather than having their names on a paper with a fundamental flaw.

13
  • 11
    I think this is the best advice you will get. You have to "fess up" and do it over. Science is hard and not all tries lead to success. If it were easy, then no one would want to pay us so handsomely to do it. Research is fundamentally looking into the unknown.
    – Buffy
    Jun 29, 2021 at 19:16
  • 13
    "So that's a big win" - excellently put. The error didn't make it into publication. Note that Michael Atiyah, one of the mathematics greats of our time offered a flawed proof of Riemann. It happens to the best of us. If you can fix it by just one month more work, that's another win, not every error can be fixed that way. Jun 29, 2021 at 19:20
  • 14
    @Captain: Let me (in part because of my deep esteem for Atiyah) suggest a more purely inspirational story: Andrew Wiles first announced a proof of Fermat's Last Theorem in 1993. This proof was (valuable but) wrong. In 1994 he fixed it. Thus one of the most lauded mathematical achievements of our lifetime was done wrong before it was done right. Jun 29, 2021 at 19:42
  • 13
    @PeteL.Clark Indeed, good example. It was not at all out of disrespect to Atiyah, but to demonstrate to OP that errors are possible and part of the endeavour. Einstein had several wrong versions of general relativity before he got it right. We do not know of what category OP's errors were, but integrity, honesty and trustworthiness is the highest capital of a scientist, and fixing a recognized error is part of that. Jun 29, 2021 at 22:24
  • 10
    Sadly I've met many researchers of the mind that it's better to find the error after publication. This way they get to publish two papers instead of one: one paper with the error, and one paper to fix it.
    – Stef
    Jun 30, 2021 at 10:02
16

This happens, and you just have to deal with it. A year and a half ago, I discovered a mistake I had made only after spending approximately 15 CPU years on computations. It led to a delay of about 6 months in the paper, during which every core of my collection of computers was busy recomputing the statistics I needed -- but at least I had confidence in the correctness of the submitted material.

In practice, the delay is often not terrible because only the computer is working on re-doing all of the calculations. While it's doing that, you can focus on the next research project, and once you have the new numbers, whatever little work is necessary to adjust the tables and graphs in the paper. In other words, little work is wasted, just time.

6

I have made plenty of mistakes in my career. Some real doozies. And I have been witness to many mistakes made by colleagues. I have found that confessing to the mistake and accepting blame and responsibility turns out the best. Even in cases where blame could be shared if you accept blame, apologize, and be honest it works out better. People will tend to attack someone who is defensive and tries to blame others. That's when they really pile on. Sure they will be frustrated and disappointed. But if they see that you feel bad and are working to fix it, they will have more sympathy and be more on your side.

5

There are already several good answers to the broader question, I just wanted to focus on the run-time aspect:

The work cannot be published unless I correct the mistakes and run the simulations again which would take at least a month more.

I was asked submit the manuscript for review by the advisor and other collaborators by next week. However, now I can't write the paper with the errors. And I have to rerun the entire work again.

Not knowing the nature of your work, I have no way of knowing whether this is an impractical suggestion, but - is it possible to reduce the rerun time by parallelising your workflow? For instance, by asking your colleagues to run some of the simulations on their own computers, or renting some cloud computing time?

1
  • @carterjack, the community at stackoverflow would be eager to help optimize and speedup your processing. In addition to GeoffreyBrent's suggestions, a few broad ideas for your consideration are profiling your code execution, reducing memory overhead, optimizing loops and functions, and minimizing disk reads/writes.
    – M.Viking
    Jul 2, 2021 at 14:17
0

There's a silver lining to your dark cloud, but only if you 'fess up: by admitting your mistake you are announcing to the world that you can be trusted and are a person of integrity, even under potentially embarrasing conditions.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .