Hacker News Reader

Comment by philipkglass | Hacker News Reader

Remarkable as they are, today’s computer chips cannot harness the quantum effects needed to solve the really important problems that quantum computers will. To solve problems that address major global challenges – like climate change or complex diseases like cancer – it’s generally accepted we will need millions of qubits working in tandem.

Are there non-classical algorithms with theoretical advantages for biological or climate simulations? Or is this just more hype-nonsense courtesy of the press office?

View on HN

In drug discovery, my field, we do a lot of quantum mechanics calculations, which are formally exponential-scaling. Using a lot of approximations, we reduce this to a polynomial, at the cost of accuracy. But a quantum computer could do the original exponential problem with polynomial scaling (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2596249). Quantum superiority for this problem would require a few hundred qubits, make drug screening a lot more accurate, and make obsolete an enormous body of theory devoted to classical approximations of this problem.

This is not my area but I understand that material sciences see the same kind of potential for QCs - predicting how new alloys or ceramics will perform by accurately simulating atomic interactions.

Yes, that is true! During my PhD program I used the same methods to study metalorganic solar cells.

If anything, fast quantum calculations would make an even bigger difference in materials science than in drug discovery. That's because druglike molecules have been studied a lot, so classical approximations (though limited) at least exist.

"Nature isn't classical, dammit, and if you want to make a simulation of nature, you'd better make it quantum mechanical, and by golly it's a wonderful problem, because it doesn't look so easy." -- Richard Feynman

"Simulating Physics with Computers", International Journal of Theoretical Physics, volume 21, 1982, p. 467-488, at p. 486 (final words) [0]

[0] https://en.wikiquote.org/wiki/Richard_Feynman#Quotes

The ability to simulate small molecules and reactions at a quantum level will be truly profound.

The limiting step in being able to manufacture any small organic molecule from raw materials is algorithmic. Imagine a molecule printer that can compile a set of synthesis steps for an arbitrary molecule via simulation. It's not as far-fetched if we can simulate hypothetical reactions easily: http://www.nature.com/news/organic-synthesis-the-robo-chemis...

somehow i don't think this is going to make drugs cheaper, the development pipeline will still take years, cost billions, and costs hundreds/thousands $ per year per consumer ..and probably won't cure anything but just be another lifelong pill

I actually agree with this, more or less. Finding promising molecules is only step 1, and drug development would still be expensive even if molecule discovery were easy.

The other parts of what you're saying are true but don't depend on what happens in quantum computing, they're just part of the for-profit drug industry. There are government and philanthropic drug hunters today who are doing good work, and if they had quantum drug discovery, they would be able to do even more.

The idea of quantum computing arose out of an attempt to make a better quantum mechanical simulator, which would allow for more efficient large molecule energy analysis, directly applicable to molecular biology.

The new computing model was later found to be generally useful and applicable to things like climate modeling too.

The simplistic intuition is that there are a lot of exact algorithms which are exponential, so we use clumsy polynomial approximations instead. A quantum computer makes many categories of exponential algorithms polynomial and therefore computable if a big and fast enough quantum computer can be built.

I haven't pinned down the research yet, but I've read in popsci mentions and arxiv abstracts that at least a couple other people have cited, that training neural nets would be accelerated by a full-scale (i.e. mega-qubit) quantum computer.

Anyone with more expertise care to point me at a relevant paper?

There has been some effort to use quantum computers to solve machine learning problems. One for training neural networks [0]; and another for unsupervised machine learning [1].

[0]: https://arxiv.org/pdf/1712.05304.pdf

[1]: https://arxiv.org/pdf/1712.05771.pdf

Here's a ppt deck covering a relevant overview of the subject.

https://medium.com/@_NicT_/quantum-machine-learning-c5ab31f6...

I can't speak to a claim that there are proven theoretical advantages for biological or climate simulations; but researchers are looking for them.

With respect to biological simulations, researchers at IBM were able to implement artificial life algorithms on a quantum computer [0]. They discuss some potential advantages to simulating life on a quantum computer versus a classical one.

Researchers have also been trying to figure out if quantum computers can be used for weather predictions [1].

[0]: https://arxiv.org/abs/1711.09442

[1]: https://link.springer.com/article/10.3103/S1068373917090011

With respect to biological simulations, researchers at IBM were able to implement artificial life algorithms on a quantum computer [0]

Genetic algorithms are similar in some regards to simulated annealing. Apparently, D-Wave's approach is kind of like quantum simulated annealing.

https://arstechnica.com/science/2017/01/explaining-the-upsid...

Protein folding and fluid simulations theoretically rely on these sorts of quantum calculations. I don't think any directly relevant algorithms have been developed because the hardware deosn't exist for it in the first place.

Good point about protein folding. Right now it's simulated with classical physics on conventional computers; quantum chemistry algorithms on classical computers scale too poorly for use with macromolecules. But a many-qubit computer could be excellent for that and other molecular-scale calculations. (Not that getting good protein folding models would cure cancer, but it's reasonably likely to aid biological research.)

Ab-initio quantum chemistry calculations certainly could scale much better on quantum computers -- I skimmed a paper once, which made it sound not easy but definitely worth working on -- and that's bound to help with bioscience once they're big enough. I'm surprised at protein folding or fluid dynamics, though -- what's the general idea on how it'd help with those?

Added: that is, I thought protein conformation energy was pretty well approximated with classical molecular-mechanics models, except during actual chemical reactions, and the difficulty was the sheer amount of time it takes this huge molecule to settle down to an energy minimum that's only a little below that of many alternative conformations. Is the biggest problem rather that we can't efficiently compute a consistent-enough approximation to the energy of a given conformation, so we can't distinguish the best conformation of the ones we simulated?

Based on what I read after I saw the protein folding suggestion, the idea is that with a Sufficiently Advanced quantum computer, you could use fully quantum calculations for small molecules and proteins alike. I'm puzzled about fluid dynamics too, though.

That sounds super-ambitious, but if humans can make LIGO, there must be a lot we can do.

All current fluid simulations are iterative and assume the fluid is a continuous indivisible mass (e.g. that it is not composed of atoms or molecules) that we then turn into discrete regions to compute the simulations. Even without common simplifying assumptions (to gain performance), this still has limits. Theoretically, simulating the regions with a quantum computer would allows us to remove these limits, and get the performance without making simplifying assumptions.

Sounds like a lot of the proponents cite improvements in machine learning, which if powerful enough might be able to make improvements.

Isn't climate simulation in and of itself a quantum calculation?

Are you thinking of chaotic systems? The time and length scales for climate simulations are such that it's one of the most-classical simulation domains I can imagine.

how can we know...

Shor's algorithm for integer factorization predates any quantum computers capable enough to do integer factorization faster than classical computers. I'm wondering if there are quantum-computer algorithms for improved biological/climate simulations that, likewise, just await large scale practical quantum computers before they become useful.

The same way Church and Turing knew what classical computers were capable of before the first computer was ever built.

The first computers actually predate them.

Reply on news.ycombinator.com