Stories
Slash Boxes
Comments

SoylentNews is people

posted by Fnord666 on Tuesday February 25 2020, @10:54AM   Printer-friendly
from the linear-is-good dept.

10,000 times faster calculations of many-body quantum dynamics possible:

A team led by Professor Michael Bonitz from the Institute of Theoretical Physics and Astrophysics at Kiel University (CAU) has now succeeded in developing a simulation method, which enables quantum mechanical calculations up to around 10,000 times faster than previously possible. They have published their findings in the current issue of the renowned scientific journal Physical Review Letters.

The new procedure of the Kiel researchers is based on one of the currently most powerful and versatile simulation techniques for quantum mechanical many-body systems.

It uses the method of so-called nonequilibrium Green functions: this allows movements and complex interactions of electrons to be described with very high accuracy, even for an extended period.

[...] to date this method is very computer-intensive: in order to predict the development of the quantum system over a ten times longer period, a computer requires a thousand times more processing time.

With the mathematical trick of introducing an additional auxiliary variable, the physicists at the CAU have now succeeded in reformulating the primary equations of nonequilibrium Green functions such that the calculation time only increases linearly with the process duration.

[...] The new calculation model of the Kiel research team not only saves expensive computing time, but also allows for simulations, which have previously been completely impossible. "We were surprised ourselves that this dramatic acceleration can also be demonstrated in practical applications," explained Bonitz.

Original publication:
Niclas Schlünzen, Jan-Philip Joost, Michael Bonitz, Achieving the Scaling Limit for Nonequilibrium Green Functions Simulations, Physical Review Letters 124, 7, (2020) DOI: 10.1103/PhysRevLett.124.076601
https://link.aps.org/doi/10.1103/PhysRevLett.124.076601


Original Submission

This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 2, Informative) by Anonymous Coward on Tuesday February 25 2020, @11:27AM (4 children)

    by Anonymous Coward on Tuesday February 25 2020, @11:27AM (#962319)

    Reading between the journalist's lines of incompetence ("10,000 times as fast"), the scientists managed to take an algorithm with a computational complexity of O(n^3) down to O(n). That is no mean feat! And a great boon to all users, surely.

    • (Score: 0) by Anonymous Coward on Tuesday February 25 2020, @11:44AM (2 children)

      by Anonymous Coward on Tuesday February 25 2020, @11:44AM (#962325)

      How is the journalist's description inaccurate?

      • (Score: 2) by Common Joe on Tuesday February 25 2020, @11:55AM (1 child)

        by Common Joe (33) <{common.joe.0101} {at} {gmail.com}> on Tuesday February 25 2020, @11:55AM (#962326) Journal

        The grandparent already explained that -- it's going from exponential down to linear computation. His information seems to come from this line:

        Thus, a ten times longer prediction period only requires ten times more computing time. In comparison with the previously-used methods, the physicists achieved an acceleration factor of approximately 10,000. This factor increases further for longer processes. Since the new approach combines two Green functions for the first time, it is called the “G1-G2 method”.

        Thus, I agree with grandparent. It seems it goes from O(n^3) down to O(n).

        • (Score: 0) by Anonymous Coward on Tuesday February 25 2020, @01:27PM

          by Anonymous Coward on Tuesday February 25 2020, @01:27PM (#962341)

          nitpick. it's polynomial in both cases. 3rd order polynomial decreases to 1st order polynomial.
          exponential would be O(2^n) (or some other basis).

    • (Score: 0) by Anonymous Coward on Tuesday February 25 2020, @03:17PM

      by Anonymous Coward on Tuesday February 25 2020, @03:17PM (#962382)

      Really good stuff. Though it looks like they're building on an improvement of O(n^2) from the original. It should be noted these aren't exact solutions and there appear to be (acceptable) accuracy tradeoffs, but still huge wins.

  • (Score: 1, Troll) by pkrasimirov on Tuesday February 25 2020, @12:12PM

    by pkrasimirov (3358) Subscriber Badge on Tuesday February 25 2020, @12:12PM (#962328)

    Break trough the discrete logarithm problem, therefore assymetric encryption, therefore SSL, therefore encrypted data and digital signatures. Combine with Deepfake audio and video. Stir with coronaviruses. Mod me troll.

  • (Score: 2) by hendrikboom on Tuesday February 25 2020, @01:10PM

    by hendrikboom (1125) Subscriber Badge on Tuesday February 25 2020, @01:10PM (#962333) Homepage Journal

    I guess this is this century's mathematics. I remember last century, a mathematician studying string theory said that string theory was next century's mathematics. Well, it's the next century now. I'm looking forward to more of this.

(1)