Polynomial Time and Extravagant Models. (By Leonid A. Levin)

Below is Section 2 of my article The Tale of One-Way Functions in Probl.Inf.Transm. 39(1), 2003.

2.1. The Downfall of RSA

This development [RSA and other applications of one-way functions] was all the more remarkable as the very existence of one-way (i.e., easy to compute, infeasible to invert) functions remains unproven and subject to repeated assaults. The first came from Shamir himself, one of the inventors of the RSA system. He proved in [Inf.Process.Lett. 8(1) 1979] that factoring (on infeasibility of which RSA depends) can be done in polynomial number of arithmetic operations. This result uses a so-called "unit-cost model," which charges one unit for each arithmetic operation, however long the operands. Squaring a number doubles its length, repeated squaring brings it quickly to cosmological sizes. Embedding a huge array of ordinary numbers into such a long one allows one arithmetic step to do much work, e.g., to check exponentially many factor candidates. The closed-minded cryptographers, however, were not convinced and this result brought a dismissal of the unit-cost model, not RSA.

Another, not dissimilar, attack is raging this very moment. It started with the brilliant result of Peter Shor [SIComp, 26(5) 1997]. He factors integers in polynomial time using an imaginary analog device, Quantum Computer (QC), inspired by the laws of quantum physics taken to their formal extreme.

2.2. Quantum Computers

QC has n interacting elements, called q-bits. A pure state of each is a unit vector on the complex plane C2. Its two components are quantum amplitudes of its two Boolean values. A state of the entire machine is a vector in the tensor product of n planes. Its 2n coordinate vectors are tensor-products of q-bit basis states, one for each n-bit combination. The machine is cooled, isolated from the environment nearly perfectly, and initialized in one of its basis states representing the input and empty memory bits. The computation is arranged as a sequence of perfectly reversible interactions of the q-bits, putting their combination in the superposition of a rapidly increasing number of basis states, each having an exponentially small amplitude. The environment may intervene with errors; the computation is done in an error-correcting way, immune to such errors as long as they are few and of special restricted forms. Otherwise, the equations of Quantum Mechanics are obeyed with unlimited precision. This is crucial since the amplitudes are exponentially small and deviations in remote (hundredth or even much further) decimal places would overwhelm the content completely. Peter Shor shows such computers capable of factoring in polynomial time. The exponentially many coordinates of their states can, using a rough analogy, explore one potential solution each and concentrate the amplitudes in the one that works.

2.3. Small Difficulties

There are many problems with such QCs. For instance, thermal isolation cannot be perfect. Tiny backgrounds of neutrinos, gravitational waves, and other exotics, cannot be shielded. Their effects on quantum amplitudes need not satisfy the restrictions on which error-correcting tools depend. Moreover, nondissipative computing gates, even classical, remain a speculation. Decades past, their existence was cheerfully proclaimed and even proved for worlds where the laws of physical interaction can be custom-designed. In our world, where the electromagnetic interaction between electrons, nuclei, and photons is about the only one readily available, circuits producing less entropy than computing remain hypothetical. So, low temperatures have limits and even a tiny amount of heat can cause severe decoherence problems. Furthermore, the uncontrollable degrees of freedom need not behave simply as heat. Interaction with the intricately correlated q-bits may put them in devilish states capable of conspiracies which defy imagination.

2.4. Remote Decimals

All such problems, however, are peanuts. The major problem is the requirement that basic quantum equations hold to multi-hundredth if not millionth decimal positions where the significant digits of the relevant quantum amplitudes reside. We have never seen a physical law valid to over a dozen decimals. Typically, every few new decimal places require major rethinking of most basic concepts. Are quantum amplitudes still complex numbers to such accuracies or do they become quaternions, colored graphs, or sick-humored gremlins? I suspect physicists would doubt even the laws of arithmetic pushed that far. In fact, we know that the most basic laws cannot all be correct to hundreds of decimals: this is where they stop being consistent with each other!

And what is the physical meaning of 500 digit long numbers? What could one possibly mean by saying "This box has a remarkable property: its q-bits contain the Ten Commandments with the amplitude whose first 500 decimal places end with 666"? What physical interpretation could this statement have even for just this one amplitude?

[Remark (not in the article): Subtleties and paradoxes of probability lead to widespread confusions. I heard questions about why similar issues do not plague tiny (albeit exponentially larger than QC amplitudes :-) probabilities. Exponentially small probabilities themselves do not have physical meaning. Their logarithms, i.e. entropies, do. Entropies are falsifiable e.g., by finding, for an experimental result, a code with significantly fewer bits than its entropy under the predicted distribution. But this (analyzed by Kolmogorov) link with the (observable) complexity, only works because probabilities are always positive (or 0). Quantum amplitude have no such property, so cannot carry this meaning.]

Close to the tensor product basis, one might have opportunities to restate the assertions using several short measurable numbers instead of one long. Such opportunities may also exist for large systems, such as lasers or condensates, where individual states matter little. But QC factoring uses amplitudes of an exponential number of highly individualized basis states. I doubt anything short of the most generic and direct use of these huge precisions would be easy to substitute. One can make the amplitudes more "physical" by choosing a less physical basis. Let us look into this.

2.5. Too Small Universe

QC proponents often say they win either way, by making a working QC or by finding a correction to Quantum Mechanics; e.g., in 2000/03/22 sci.physics.research usenet article "Re: When will quantum computers become practical?" http://groups.google.com/group/sci.physics.research/msg/dc191983dfa1477b, Peter Shor said: "If there are nonlinearities in quantum mechanics which are detectable by watching quantum computers fail, physicists will be VERY interested (I would expect a Nobel prize for conclusive evidence of this)."

Consider, however, this scenario. With few q-bits, QC is eventually made to work. The progress stops, though, long before QC factoring starts competing with pencils. The QC people then demand some noble prize for the correction to the Quantum Mechanics. But the committee wants more specifics than simply a nonworking machine, so something like observing the state of the QC is needed. Then they find the Universe too small for observing individual states of the needed dimensions and accuracy. (Raising sufficient funds to compete with pencil factoring may justify a Nobel Prize in Economics.)

Let us make some calculations. In cryptography, the length n of the integers to factor may be a thousand bits (and could easily be millions.) By *n, I will mean a reasonable power of n. A 2*n-dimensional space H has 2exp(*n) nearly orthogonal vectors. Take a generic v in H. The minimal size of a machine which can recognize or generate v (approximately) is K=exp(*n) --- far larger than our Universe. This comes from a cardinality argument: 2*K machines of K atoms. Let us call such v "megastates."

There is a big difference between untested and untestable regimes. Claims about individual megastates are untestable. I can imagine a feasible way to separate any two QC states from each other. Yet, as this calculation shows, no machine can separate a generic QC state from the set of all states more distant from it than QC tolerates. So, what thought experiments can probe the QC to be in the state described with the accuracy needed? I would allow to use the resources of the entire Universe, but not more!

Archimedes made a great discovery that digital representation of numbers is exponentially more efficient than analog ones (sand pile sizes). Many subsequent analog devices yielded unimpressive results. It is not clear why QCs should be an exception.

2.6. Metric versus Topology

A gap in quantum formalism may be contributing to the confusion. Approximation has two subtly different aspects: metric and topology. Metric tells how close our ideal point is to a specific wrong one. Topology tells how close it is to the combination of all unacceptable (non-neighboring) points. This may differ from the distance to the closest unacceptable point, especially for quantum systems.

In infinite dimensions, the distinction between 0 and positive separation between a point and a set varies with topologies. In finite dimensions, 0-vs.-positive distinction is too coarse: all topologies agree. Since 2500 is finite only in a very philosophical sense, one needs a quantitative refinement, some sort of a weak-topological (not metric) depth of a neighborhood polynomially related to resources required for precision to a given depth. Then, precision to reasonable depths would be physical, e.g., allow one to generate points inside the neighborhood, distinguish its center from the outside, etc.

Metric defines e-neighborhoods and is richer in that than topology where the specific value of e is lost (only e > 0 is assured). However, metric is restricted by the axiom that the intersection of any set of e-neighborhoods is always another e-neighborhood. Quantum proximity may require both benefits: defined depth e and freedom to express it by formulas violating the "intersection axiom." Here is an example of such a violation, without pretense of relevance to our needs. Suppose a neighborhood of 0 is given by a set of linear inequalities fi(x)<1; then its depth may be taken as 1/sumi || fi||. Restricting x to the unit sphere would render this depth quadratically close to metric depth. A more relevant formula may need preferred treatment of tensor product basis.

2.7. The Cheaper Boon

QC of the sort that factors long numbers seems firmly rooted in science fiction. It is a pity that popular accounts do not distinguish it from much more believable ideas, like Quantum Cryptography, Quantum Communications, or the sort of Quantum Computing that deals primarily with locality restrictions, such as fast search of long arrays. It is worth noting that the reasons why QC must fail are by no means clear; they merit thorough investigation. The answer may bring much greater benefits in the understanding of basic physical concepts than any benefits factoring devices could promise. The present attitude is analogous to, say, Maxwell selling the Daemon of his famous thought experiment as a path to cheaper electricity from heat. If he did, much of insights of today's thermodynamics might be lost or delayed.

The rest of the article ignores any extravagant models and stands fully committed to the Polynomial Overhead Church--Turing Thesis: Any computation that takes t steps on an s-bit device can be simulated by a Turing Machine in sO(1)t steps within sO(1) cells.