Quantum Duel: Gil Kalai × Matthias Christandl (A recording of the debate)

Do quantum computers exist? Is there any hope that we will ever be able to build relevant quantum computers in the future? These questions were the subject of a friendly debate between two leading global experts on May 20, 2025, at the National Library of Technology in Prague. A recording of their discussion is available on YouTube and on the SlidesLive platform.

The duel took place on Tuesday, May 20, 2025, in the Balling Hall of the National Library of Technology in Prague.

• Matthias Christandl, University of Copenhagen
• Gil Kalai, Reichman University and Hebrew University of Jerusalem

Matthias Christandl argues that qubits (units of information that are “both 0 and 1 at the same time”) have already been created, and that this will lead to the development of useful quantum computers. Gil Kalai does not dispute the creation of qubits, but he argues that, because of their inherent properties, we will never have practically relevant quantum computers:

• SlidesLive
• YouTube

After the duel, Gil Kalai provided written answers to questions from the audience that there had not been time to address during the event:

• Novice, For Gil: From what I understand, you believe this is an engineering challenge and not exactly of physics (theoretical, to be precise). If yes, should not the debate involve a quantum engineer?
  Gil: I posit that there are inherent physical obstacles to reducing the error level. The obstacle represents a theoretical analysis of quantum computers on the intermediate scale and the probability distributions they produce.  So, if I am correct, it is not “just” an engineering challenge. Adding quantum engineers to the discussion is always welcome. In one assertion, I do rely on insights coming from experimentalists and quantum engineers: that error rates required for creating good quality quantum error correcting codes are even lower than those required for demonstrating “quantum computational supremacy”.
• Anonymous: Could you both debate more about where your models of quantum computers differ? How is it possible that you reach different conclusions?
  Gil: As far as I can see, our models of noisy quantum computers are the same. I argue that reducing the error rates will hit a barrier. Regarding the second question, it is not uncommon that scientists reach different conclusions from the available theoretical and experimental information. 
• FKM: Two qubit gates always involve non-computational states, introducing a tradeoff between speed, leakage and errors (when "fixing" the leakage). That might be a microscopic origin of irreducible noise. Do we have definite knowledge about impact of non-computational states on the error rates?
  Gil: The threshold theorem deals with arbitrary forms of errors for two-qubit gates under the condition that the error rate is sufficiently small.
• Dmitry Grinko, To Gil: in your answer to Ronald you mentioned that LDP of NISQ is not an obstacle to harder classical computation. Why may it also NOT be an obstacle to fault-tolerant quantum computation? What is your precise argument for why easiness of NISQ implies impossibility of quantum fault-tolerance?
  Gil: the mechanism for creating robust classical information (repetition plus majority) is supported by the class LDP. In contrast, to the best of our knowledge, quantum error correction is not supported by this class. 
• Jiri Kolafa: If we combine n physical qubits into 1 logical qubit, how does the noise reduce? As poly(1/n) or exp(-kn)? In the 2nd case, I believe q computing...
  Gil: For distance-d surface codes, n behaves like d^2 and the noise scale like exp (-kd) (which is exp (-k sqrt n). (So, it comes close to your 2nd case, Jiri.)
• Anonymous: Do I understand correctly, that the claim that NISQ devices are limited to some computational class is based on experimental results from NISQ devices? / Does this necessarily apply to all possible quantum hardware? / Why would this imply that a fault-tolerant computer couldn't be built at all?
  Gil: No, it is based on theoretical considerations. / Yes. / Because fault-tolerant computers on the large scale require building good-quality quantum error correction in the intermediate (NISQ) scale. And this is not supported by the limited computational class describing the intermediate scale. 
• Anonymous: Prof. Kalai, I understand your argument as follows: because of an assumed complexity barrier, there is an engineering barrier. Is it correct? If so, assuming that fault-tolerant QC are built, will it imply fundamental progress in complexity theory?
  Gil: Yes. If fault-tolerant QC are built this would imply that my deduction of engineering barriers from (asymptotic) computational behavior is incorrect.
• Michal Sustr: A good theory makes prediction about problems people care about. Prof. Kalai, what is the prediction of required noise error of individual physical qubit based on problem size? For example Shor’s algorithm for given # of input bits.
  Gil: The threshold theorem asserts that (under certain conditions) the required error level for individual physical qubits (and gates) to allow Shor’s algorithm is a certain constant (that does not depend on the # of input bits). The required level may depend on various properties of the noise. (Such as long-range correlations.)
  My argument gives a firm prediction that reaching “quantum supremacy” and very good quality quantum error correction codes is beyond reach. Like other applications of asymptotic computational complexity insights, it does not give a firm prediction on the barrier.
  In the debate, Matthias asked me to give a firm prediction of the level of improvement that cannot be reached, and my response was that I expect that improvements of the quality of physical qubits and gates by two orders of magnitude is already beyond reach. 
• Anonymous: Large parts of computational complexity are based on certain computational complexity hypotheses. Isn’t it too brave to dismiss existence of quantum computers based on something we are not sure is actually correct?
  Gil: This is correct in general, but the computational class I consider is so low-low-level that its weakness does not depend on unproven hypotheses. 
• Michal Sustr: Prof. Kalai, why did you find Majorana Zero Modes unconvincing?
  Gil: On this issue, I have secondhand knowledge: the experimental claims for MZM since 2012 may represent, rather than MZM, more mundane quantum states called “Andreev’s quasi particles”. As far as I know, the referees of the current Microsoft paper found the evidence for MZM unconvincing and decided to publish the paper for other advances reported in the paper. (The authors still believe that they created MZMs.) In these circumstances, I regard the claims unconvincing.