6 ms·

Why does this happen so much? I mean, this is science: shouldn’t chemicals react as they do in theory? It doesn’t seem like it’d leave a lot to chance.

quanticle 7y ago
Computational chemistry involves solving multiple NP-hard problems. There is an increasing reliance on computational models and simulation, but chemists realize that the models are approximations (since it's computationally infeasible to have perfect fidelity simulations). The only surefire way to observe a compound's properties is to synthesize it and measure.
- OscarCunningham 7y ago
  Surely there can't be any problems in computational chemistry that have actually been proven to be NP-hard? Simulating an experiment atom-for-atom is in BQP, which is not known to contain NP (and is suspected not to contain NP). EDIT: To put it another way, we can't solve NP problems by doing chemistry experiments.
  - 0xffff2 7y ago
    Surely you would need to be modelling at a subatomic level to accurately simulate chemical interactions?
    - OscarCunningham 7y ago
      Sure. It's not an essential distinction though, because subatomic physics is also simulable in BQP.
      - selimthegrim 7y ago
        N-representability?
  - the8472 7y ago
    Layman understanding ahead, see scott aaronson's NP-complete Problems and Physical Reality[0] if you want more rigorous arguments. BQP gets you collapsing measurements. The Schrödinger equation (NP-hard) gives you all probabilities at once, solving it is equivalent to performing non-collapsing measurements. [0] https://www.scottaaronson.com/papers/npcomplete.pdf https://www.scottaaronson.com/papers/npcomplete.pdf
  - jhrmnn 7y ago
    Here’s one example. Though it’s probably different from what you had in mind https://arxiv.org/abs/cond-mat/0408370 https://arxiv.org/abs/cond-mat/0408370
inflatableDodo 7y ago
In the words of the Great Yogi (Berra); 'In theory there is no difference between theory and practice. In practice there is.'
OscarCunningham 7y ago
It's often impossible to theoretically predict how two chemicals will react. The interaction of the molecules is governed by quantum mechanics, which can't be simulated quickly by classical computers.
nabla9 7y ago
Chemistry never had good theoretical classical theory like physics had. Theories have been just rules of thumb or heuristics based on experiments. Theoretically all chemistry is quantum physics with nucleons, electrons and photons (quantum chemistry). You need to numerically simulate what happens in reactions and it's very expensive.
- derefr 7y ago
  Predicting reactions is hard. But why is predicting the physical—rather than chemical—properties of simple inorganic molecules like this one hard? Presumably, trying to figure out how light interacts with a molecule is an easier problem than, say, protein folding, no?
  - saagarjha 7y ago
    No, because there’s a bunch of quantum goop that gets really complicated and your eyes not going to be able to replicate your theoretical 100% pure result anyways.
  - rout39574 7y ago
    What does properties mean, to you? For the sweetness test, the property is the way the compound (or its breakdown constituents, or the compounds whose formation it catalyzes) reacts to our bodies. So you're trying to figure out how our tastebuds work, in response to a novel stimulus.
    - derefr 7y ago
      The fact of whether or not a chemical reaction takes place in a certain context is a chemical property of a substance. Any fact about a substance on its own, not interacting with other molecules (or at least interacting with them a way that causes no chemical reactions to take place, e.g. being cooled by convection of Neon gas, or scratching in a Moh’s hardness test, or piezoelectric effects), is a physical property. Sweetness is, as you’ve defined it, a chemical property, so it’s not one of the things I was talking about being “easy” to predict. But pigmentation is a physical property, as it is a reaction between a molecule and inbound photons†. Likewise, the output of a “pigmentation calculation”—if we know of one—should just be more photons. Wavefunction in, wavefunction out. We know how photons work. We know how our eyes perceive them. As long as we have a framework for figuring out how a material physically manipulates the photons absorbed by it, we should be able to calculate the pigmentation of every simple molecule by brute force. —— † The pigmented molecule can be in the chemical form or physical state that gives it that pigment very ephemerally, due to e.g. a clock reaction. But at any given instant in time, we can say “the molecule is in this state, so what function does it apply to inbound photons now.” Note also that I’m defining pigmentation here, not total appearance, so this ignores any fluorescence, phosphorescence, triboluminescence, Cherenkov radiation, etc. that the molecule also emits and which we perceive as light. All those purely-physical processes are able to be admitted into this model without actually making it untenable, but it becomes harder to understand, since it becomes a Feynman diagram thing, rather than simple transformation over wavefunctions.

quanticle 7y ago

Computational chemistry involves solving multiple NP-hard problems. There is an increasing reliance on computational models and simulation, but chemists realize that the models are approximations (since it's computationally infeasible to have perfect fidelity simulations). The only surefire way to observe a compound's properties is to synthesize it and measure.

OscarCunningham 7y ago
Surely there can't be any problems in computational chemistry that have actually been proven to be NP-hard? Simulating an experiment atom-for-atom is in BQP, which is not known to contain NP (and is suspected not to contain NP). EDIT: To put it another way, we can't solve NP problems by doing chemistry experiments.
- 0xffff2 7y ago
  Surely you would need to be modelling at a subatomic level to accurately simulate chemical interactions?
  - OscarCunningham 7y ago
    Sure. It's not an essential distinction though, because subatomic physics is also simulable in BQP.
    - selimthegrim 7y ago
      N-representability?
- the8472 7y ago
  Layman understanding ahead, see scott aaronson's NP-complete Problems and Physical Reality[0] if you want more rigorous arguments. BQP gets you collapsing measurements. The Schrödinger equation (NP-hard) gives you all probabilities at once, solving it is equivalent to performing non-collapsing measurements. [0] https://www.scottaaronson.com/papers/npcomplete.pdf https://www.scottaaronson.com/papers/npcomplete.pdf
- jhrmnn 7y ago
  Here’s one example. Though it’s probably different from what you had in mind https://arxiv.org/abs/cond-mat/0408370 https://arxiv.org/abs/cond-mat/0408370

inflatableDodo 7y ago

In the words of the Great Yogi (Berra); 'In theory there is no difference between theory and practice. In practice there is.'

OscarCunningham 7y ago

It's often impossible to theoretically predict how two chemicals will react. The interaction of the molecules is governed by quantum mechanics, which can't be simulated quickly by classical computers.

nabla9 7y ago

Chemistry never had good theoretical classical theory like physics had. Theories have been just rules of thumb or heuristics based on experiments. Theoretically all chemistry is quantum physics with nucleons, electrons and photons (quantum chemistry). You need to numerically simulate what happens in reactions and it's very expensive.

derefr 7y ago
Predicting reactions is hard. But why is predicting the physical—rather than chemical—properties of simple inorganic molecules like this one hard? Presumably, trying to figure out how light interacts with a molecule is an easier problem than, say, protein folding, no?
- saagarjha 7y ago
  No, because there’s a bunch of quantum goop that gets really complicated and your eyes not going to be able to replicate your theoretical 100% pure result anyways.
- rout39574 7y ago
  What does properties mean, to you? For the sweetness test, the property is the way the compound (or its breakdown constituents, or the compounds whose formation it catalyzes) reacts to our bodies. So you're trying to figure out how our tastebuds work, in response to a novel stimulus.
  - derefr 7y ago
    The fact of whether or not a chemical reaction takes place in a certain context is a chemical property of a substance. Any fact about a substance on its own, not interacting with other molecules (or at least interacting with them a way that causes no chemical reactions to take place, e.g. being cooled by convection of Neon gas, or scratching in a Moh’s hardness test, or piezoelectric effects), is a physical property. Sweetness is, as you’ve defined it, a chemical property, so it’s not one of the things I was talking about being “easy” to predict. But pigmentation is a physical property, as it is a reaction between a molecule and inbound photons†. Likewise, the output of a “pigmentation calculation”—if we know of one—should just be more photons. Wavefunction in, wavefunction out. We know how photons work. We know how our eyes perceive them. As long as we have a framework for figuring out how a material physically manipulates the photons absorbed by it, we should be able to calculate the pigmentation of every simple molecule by brute force. —— † The pigmented molecule can be in the chemical form or physical state that gives it that pigment very ephemerally, due to e.g. a clock reaction. But at any given instant in time, we can say “the molecule is in this state, so what function does it apply to inbound photons now.” Note also that I’m defining pigmentation here, not total appearance, so this ignores any fluorescence, phosphorescence, triboluminescence, Cherenkov radiation, etc. that the molecule also emits and which we perceive as light. All those purely-physical processes are able to be admitted into this model without actually making it untenable, but it becomes harder to understand, since it becomes a Feynman diagram thing, rather than simple transformation over wavefunctions.