1 of 68

Towards a Complexity Theory �for the Quantum Age

Henry Yuen

Columbia

Image credit: Dall-E

2 of 68

What is the Quantum Age?

Quantum computers are commonplace.

They compute on quantum data.

They communicate over a quantum network.

What are the computational problems of the Quantum Age?

How do we reason about their complexity?

3 of 68

Wait a minute…

…don’t we already have this?

Symposium on Theory of Computing 1993

4 of 68

30 years of Quantum Complexity Theory

  • Rich complexity classes: BQP, QMA, QMA(2), MIP*, …

  • Interesting computational problems: Hamiltonian simulation, approximating the Jones polynomial, ground energy estimation, …

  • Insights into other fields: Hamiltonian complexity and condensed matter, quantum interactive proofs and operator algebras, …

  • Foundations of quantum advantage: BosonSampling, Random Circuit Sampling, complexity-theoretic evidence for quantum supremacy, …

5 of 68

30 years of Quantum Complexity Theory

A theory of “classical input/classical output” problems

Classical

Input

Quantum

Model of Computation

Classical

Output

6 of 68

30 years of Quantum Complexity Theory

A theory of “classical input/classical output” problems

Integer N

Quantum Computer

Factors p, q

Example: Factoring

7 of 68

30 years of Quantum Complexity Theory

A theory of “classical input/classical output” problems

Local Hamiltonian H

QMA verifier-prover protocol

Ground energy ± ℇ

Example: Local Hamiltonians problem

 

8 of 68

With classical input/classical output problems, we can compare classical vs quantum models of computation.

It also allows us to connect Quantum Complexity Theory with Classical Complexity Theory.

e.g., “BosonSampling is hard for classical computers unless the polynomial hierarchy collapses”

9 of 68

Quantum in, quantum out

In the Quantum Age, quantum computers will perform tasks that

don’t make sense on classical computers.

10 of 68

Quantum in, quantum out

Local Hamiltonian H

Ground state of H

Example: Preparing complex quantum states

In the Quantum Age, quantum computers will perform tasks that

don’t make sense on classical computers.

Quantum Computer

11 of 68

Quantum in, quantum out

Noisy quantum state

Denoised state

Example: Correcting noisy quantum information

In the Quantum Age, quantum computers will perform tasks that

don’t make sense on classical computers.

Quantum Computer

12 of 68

Quantum in, quantum out

Quantum state in�unknown phase

Quantum Computer

Phase label

Example: Classifying quantum phases of matter

In the Quantum Age, quantum computers will perform tasks that

don’t make sense on classical computers.

13 of 68

Quantum in, quantum out

These are natural tasks in the Quantum Age.

Can Quantum Complexity Theory explain when they are easy?

Or when they are hard?

14 of 68

Quantum in, quantum out

These are natural tasks in the Quantum Age.

Can Quantum Complexity Theory explain when they are easy?

Or when they are hard?

Recently, Quantum Complexity Theory has started looking incomplete.

15 of 68

Quantum information 101

  •  

16 of 68

Quantum information is different

  •  

17 of 68

The story of�quantum bit commitments

18 of 68

Bit commitments

Cryptographic analogue of putting a message in a sealed envelope that is opened later. A commitment protocol involves two phases, commit and reveal.

 

Committer

Receiver

 

Hiding security: commitment c does not reveal bit b to receiver.

 

19 of 68

Bit commitments

Cryptographic analogue of putting a message in a sealed envelope that is opened later. A commitment protocol involves two phases, commit and reveal.

Reveal phase: committer opens commitment c to reveal b.

Committer

Receiver

Binding security: committer cannot open to opposite bit 1 – b.

 

 

20 of 68

Security of commitments

Commitments are fundamental, used in multiparty computation and zero-knowledge protocols.

21 of 68

Security of commitments

Commitments are fundamental, used in multiparty computation and zero-knowledge protocols.

The security of commitments requires some computational assumptions, e.g. P ≠ NP.

If both parties can solve NP-complete problems, either �Hiding security or Binding security (or both) can be broken…

22 of 68

Security of commitments

Commitments are fundamental, used in multiparty computation and zero-knowledge protocols.

The security of commitments requires some computational assumptions, e.g. P ≠ NP.

If both parties can solve NP-complete problems, either �Hiding security or Binding security (or both) can be broken… ��…at least, if the scheme is classical.

23 of 68

Quantum bit commitments

What if the scheme takes advantage of quantum physics?

Committer

Receiver

Commit phase: Committer, receiver exchange quantum messages. �Afterwards, they can be entangled.

Hiding security: receiver cannot learn b by measuring her quantum state.

24 of 68

Quantum bit commitments

What if the scheme takes advantage of quantum physics?

Committer

Receiver

Reveal phase: they exchange more quantum messages. �Afterwards, receiver measures her state to verify commitment to b.

Binding security: even if committer deviates from the protocol, �receiver accepts 1 - b with very small probability.

25 of 68

Quantum bit commitments

In 1980s, cryptographers were fascinated by the possibility of unconditionally secure quantum commitment schemes.

This was motivated by quantum key distribution (QKD), �whose security is based only “on the laws of quantum physics”, rather than unproven complexity conjectures.

Some researchers (1993): We found it!

Lo, Chau/Mayers (1996): No you didn’t.

26 of 68

Quantum bit commitments

Lo, Chau/Mayers 1996: The security of commitment schemes �(even quantum ones) must rely on computational assumptions.

27 of 68

Lo, Chau/Mayers argument

  •  

Committer

Receiver

 

 

 

28 of 68

Lo, Chau/Mayers argument

  •  

Committer

Receiver

 

 

Suppose, for contradiction, that both Hiding and Binding security do not require computational assumptions.

29 of 68

Uhlmann’s theorem

  •  

Uhlmann’s theorem is central in quantum information: it quantifies how well one can locally transform one entangled state into another.

Uhlmann transformation on A

identity transformation on B

 

30 of 68

Lo, Chau/Mayers argument

  •  

Committer

Receiver

 

 

 

31 of 68

Lo, Chau/Mayers argument

  •  

Committer

Receiver

 

 

32 of 68

Complexity of Uhlmann transformations

  •  

33 of 68

Complexity of Uhlmann transformations

Based on classical cryptography we expect Uhlmann transformations to be difficult.

Breaking commitment �schemes

 

Inverting one-way functions

 

Implementing �Uhlmann transformations

Lo, Chau / Mayers (1996)

Classical commitment�construction from OWFs (e.g. Naor 1991)

 

34 of 68

Complexity of Uhlmann transformations

What about upper bounds on their complexity?

 

NP

Implementing �Uhlmann transformations

Breaking commitment �schemes

 

 

Not clear how Uhlmann transformations reduce to classical search problems with efficient verification…

?

35 of 68

Complexity of Uhlmann transformations

What about upper bounds on their complexity?

 

Implementing �Uhlmann transformations

Breaking commitment �schemes

 

 

Something funny is going on… EXP could mean multiple things.

EXP

?

36 of 68

Complexity of Uhlmann transformations

What about upper bounds on their complexity?

 

Implementing �Uhlmann transformations

Breaking commitment �schemes

 

 

Fact: Uhlmann transformations can be implemented by a quantum computer in exponential time.

EXP, as in actually �having exponential time

37 of 68

Complexity of Uhlmann transformations

What about upper bounds on their complexity?

 

Implementing �Uhlmann transformations

Breaking commitment �schemes

 

 

Can Uhlmann transformations be efficiently implemented�with oracle access to an EXP-complete decision problem?

EXP, the class of decision �problems solvable by exponential-time Turing machines.

?

38 of 68

Complexity of Uhlmann transformations

What about upper bounds on their complexity?

 

Implementing �Uhlmann transformations

Breaking commitment �schemes

 

 

Can Uhlmann transformations be efficiently implemented�with oracle access to the Halting problem?

Halting Problem

?!

39 of 68

Unitary Synthesis Question

  •  

40 of 68

Unitary Synthesis Question

  •  

41 of 68

Unitary Synthesis Question

  •  

42 of 68

Unitary Synthesis Question

  •  

43 of 68

Quantum in, quantum out is different

No-cloning theorem, fragility of quantum information makes quantum input/quantum output problems fundamentally different from classical computation tasks.

  • Since the algorithm doesn’t have a classical description of input, it doesn’t have a classical description of output.

  • Thus oracle access to classical functions – even fantastically complex ones – doesn’t seem to help ”shortcut” a difficult quantum transformation.

44 of 68

Other evidence

No search-to-decision reductions for QMA (“quantum NP”)

    • If ground energy estimation is easy, is ground state preparation easy?
    • Evidence the answer is No.

Even if P = NP, computationally-secure quantum cryptography can still exist.

[Kretschmer ‘21], [Kretschmer, Qian, Sinha, Tal ‘23]

[Irani, Rao, Natarajan, Nirkhe, Yuen ‘21]

45 of 68

Summary so far

In the Quantum Age, there are interesting computational problems with quantum inputs and quantum outputs.

Traditional complexity theory of decision problems does not seem to be the right framework to reason about quantum cryptography.

Lack of search-to-decision, quantum cryptography in Algorithmica, and the Unitary Synthesis Question suggest we need a �“fully quantum” complexity theory for “fully quantum” computational problems.

46 of 68

Goals of a new complexity theory

  • Identify natural quantum tasks from the sciences.
    • condensed matter physics, high energy physics, chemistry, computer science, cryptography, optimization, experimental physics,…
  • Show reductions between them
  • Classify them by complexity
  • Identify complete problems
  • Prove separations
  • Use hardness for cryptography

Next: an illustration of these goals

47 of 68

Unitary Complexity �and the�Uhlmann Transformation Problem

John Bostanci

Columbia

Yuval Efron

Columbia

Alexander Poremba

MIT

Tony Metger

ETH

Luowen Qian

Boston University

Joint work with:

48 of 68

Let’s go on safari…

Unscrambling the Hawking radiation of a black hole

Quantum state compression

Decoding noisy quantum channels

Breaking quantum cryptography

Optimal prover �strategies in a quantum�interactive proof

Uhlmann Transformations

49 of 68

Unitary synthesis problems

 

 

Quantum Model of Computation

 

 

50 of 68

Unitary synthesis problems

 

Quantum Model of Computation

 

 

 

 

51 of 68

Unitary complexity theory

A unitary complexity class is a collection of unitary synthesis problems.

unitaryBQP: unitary synthesis problems implementable by polynomial-time algorithms.

unitaryPSPACE: unitary synthesis problems implementable by polynomial-space algorithms.

52 of 68

Reductions

 

Polynomial-time query algorithm

 

 

 

 

 

 

 

 

 

53 of 68

Uhlmann Transformation Problem

An example of a unitary synthesis problem.

 

 

 

 

(

,

)

 

A

B

A

B

specifies

54 of 68

How hard are Uhlmann Transformations?

  •  

Theorem* [BEMPQY ‘23]: UHLMANN is complete for unitarySZK, a unitary complexity analogue of statistical zero knowledge.

 

Is UHLMANN a complete problem for a natural complexity class?

55 of 68

Quantum commitments

Lo, Chau/Mayers: task of breaking quantum commitments polynomial-time reduces to UHLMANN.

Converse?

 

Idea: Hard instances of UHLMANN yield “weak quantum commitments” that can be amplified to “strong” commitments using the quantum parallel repetition theorem of [BQSY ‘23].

56 of 68

Decoding noisy quantum channels

Noisy Quantum Channel

Decoder

Encoder

Often a random code

Often unknown complexity

Fundamental goal in quantum Shannon theory: given a noisy quantum channel, find encoding/decoding algorithms that maximize the information recovered.

Source

Dest

57 of 68

Decoding noisy quantum channels

Noisy Quantum Channel

Decoder

Encoder

Often a random code

Often unknown complexity

Fundamental goal in quantum Shannon theory: given a noisy quantum channel, find encoding/decoding algorithms that maximize the information recovered.

Source

Dest

58 of 68

Decoding decodable quantum channels

Decodable

Channel

Decoder

Source

Dest

A channel that is �promised to have a decoder.

What is the complexity?

An algorithmic version of the problem: given circuit description of a decodable channel, implement a decoder that can recover the source with high fidelity.

59 of 68

Decoding decodable quantum channels

 

Decodable

Channel

Decoder

Source

Dest

A channel that is �promised to have a decoder.

What is the complexity?

60 of 68

The black hole information paradox

Quantum physics: all physical processes are reversible

General relativity: information thrown into a black hole is lost forever

Stephen Hawking: black holes evaporate due to radiation

Black hole

After 1060 years

Cloud of Hawking radiation

Maybe all the

information is in here?

Over the years, physicists have tried to resolve the paradox in different ways.

61 of 68

The black hole information paradox

The latest wrinkle: even if the information is preserved in the Hawking radiation, someone could potentially decode the information, jump into the black hole, and suddenly find two copies of the same quantum state.

In 2013, Harlow and Hayden proposed a complexity-theoretic way out: decoding the Hawking radiation can’t be done in polynomial time!

 

See also “Black-Hole Radiation Decoding is Quantum Cryptography” by Brakerski (2022).

62 of 68

Succinct Uhlmann

  •  

Theorem*: SUCCINCT-UHLMANN is complete for unitaryPSPACE.

Theorem*: SUCCINCT-UHLMANN is complete for unitaryQIP.

Corollary*: unitaryQIP = unitaryPSPACE.

63 of 68

Complexity of quantum cryptography

The security of quantum commitments is equivalent to hardness of UHLMANN.

But there are many other quantum crypto primitives: quantum money, copy-protection, unclonable encryption, certified deletion,… What is the complexity of breaking them in general?

Theorem*: A falsifiable quantum cryptographic assumption is either unconditionally secure, or can be efficiently broken in unitaryPSPACE.

A falsifiable crypto assumption is one that can be efficiently refuted via an interactive game.

64 of 68

Summary and�Future Directions

65 of 68

A framework for unitary complexity theory

  • Traditional complexity theory has hard time reasoning about “fully quantum” tasks: preparing ground states, breaking quantum crypto, quantum channel coding.

  • A framework of unitary synthesis problems and unitary complexity classes seems better suited to study the complexity of “fully quantum” tasks.

  • Today: Uhlmann Transformation Problem captures the complexity of:
    • Breaking quantum commitments
    • Decoding noisy channels
    • Zero-knowledge unitary synthesis
    • Harlow-Hayden black hole radiation decoding task

66 of 68

avgUnitaryPSPACE = avgUnitaryQIP

avgUnitarySZK

avgUnitaryBQP

SuccinctUhlmann

Uhlmann

Hamiltonian Evolution

Decodable Channels

Breaking Commitments

The unitary complexity�landscape so far….

Unscrambling�Hawking Radiation

Fast-Forwarding Hamiltonian Evolution

We have a much better�understanding of the average-case�versions of these classes…

Interference

Detection

Compression

67 of 68

A zoo of “fully quantum” problems

Quantum Info Theory:

Coding over noisy channels

Entanglement distillation

State compression

State redistribution

Hypothesis testing

State/process tomography

Quantum Cryptography (breaking security of):

Quantum money

Commitment schemes

Pseudorandom states

EFI pairs

One-way state generators

Quantum copy-protection

Physics:

Preparing ground states

Hamiltonian time evolution

Decoding Hawking radiation

Interference detection

Metrology

Quantum algorithms:

QSVT

Identifying coset states

Unitary property testing

68 of 68

Many directions to explore

  • A new complexity theory: Define and study other unitary complexity classes (e.g., unitaryQMA?), and the relationships between each other. Populate them with interesting quantum input/output tasks.

  • Foundations of quantum crypto: what is complexity of breaking quantum money or other uncloneable quantum primitives? Are there average-case to worst-case reductions? Is there a minimal complexity assumption needed for quantum crypto?

  • Traditional vs unitary complexity theory: unitaryPSPACE = unitaryBQPPSPACE?

Thank You For Listening!