Quantum isomorphism is equivalent to equality of homomorphism counts from planar graphs
Laura Manˇcinska1 David E. Roberson2
1QMATH, University of Copenhagen
2Technical University of Denmark
QIP 2021 February 4, 2021 2 / 23 Problem: Entanglement-assisted strategies for arbitrary • nonlocal games are hard to analyze Line of attack: Focus on a well-behaved class of games •
Nonlocal games provide a general framework for studying • entanglement Line of attack: Focus on a well-behaved class of games •
Nonlocal games provide a general framework for studying • entanglement Problem: Entanglement-assisted strategies for arbitrary • nonlocal games are hard to analyze Nonlocal games provide a general framework for studying • entanglement Problem: Entanglement-assisted strategies for arbitrary • nonlocal games are hard to analyze Line of attack: Focus on a well-behaved class of games • This talk
PART I
Quantum isomorphism and different ways to think about it: Nonlocal games • Matrix formulations • Homomorphism counts •
PART II
Elements of the proof: Intertwiners of quantum groups • Bi-labeled graphs • Homomorphism matrices •
4 / 23 A map f : V(G) V(H) is an isomorphism from G to H if → f is a bijection and • g ∼ g if and only if f(g) ∼ f(g ). • 0 0
If such a map exists, we say that G and H are isomorphic and write G =∼ H.
Matrix formulation: PAGP† = AH for some permutation matrix P
Graph isomorphism
∼=
5 / 23 If such a map exists, we say that G and H are isomorphic and write G =∼ H.
Matrix formulation: PAGP† = AH for some permutation matrix P
Graph isomorphism
∼=
A map f : V(G) V(H) is an isomorphism from G to H if → f is a bijection and • g ∼ g if and only if f(g) ∼ f(g ). • 0 0
5 / 23 Matrix formulation: PAGP† = AH for some permutation matrix P
Graph isomorphism
∼=
A map f : V(G) V(H) is an isomorphism from G to H if → f is a bijection and • g ∼ g if and only if f(g) ∼ f(g ). • 0 0
If such a map exists, we say that G and H are isomorphic and write G =∼ H.
5 / 23 Graph isomorphism
∼=
A map f : V(G) V(H) is an isomorphism from G to H if → f is a bijection and • g ∼ g if and only if f(g) ∼ f(g ). • 0 0
If such a map exists, we say that G and H are isomorphic and write G =∼ H.
Matrix formulation: PAGP† = AH for some permutation matrix P
5 / 23 To win players must reply h, h • 0 such that rel(h, h0) = rel(g, g0)
No communication during game •
Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R
A B
1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
6 / 23 To win players must reply h, h • 0 such that rel(h, h0) = rel(g, g0)
No communication during game •
Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
R
A B
1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
6 / 23 To win players must reply h, h • 0 such that rel(h, h0) = rel(g, g0)
No communication during game •
Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R g g0
A B
6 / 23 To win players must reply h, h • 0 such that rel(h, h0) = rel(g, g0)
No communication during game •
Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R g g0
A B h h0
6 / 23 No communication during game •
Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R To win players must reply h, h0 g g0 • such that rel(h, h0) = rel(g, g0)
A B h h0
6 / 23 Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R To win players must reply h, h0 g g0 • such that rel(h, h0) = rel(g, g0)
No communication during game • A B h h0
6 / 23 Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R To win players must reply h, h0 g g0 • such that rel(h, h0) = rel(g, g0)
No communication during game • A B h h0
Fact. G =∼ H Classical players can win the game with certainty ⇔
6 / 23 1We work in the commuting rather than the tensor-product model.
(G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R To win players must reply h, h0 g g0 • such that rel(h, h0) = rel(g, g0)
No communication during game • A B h h0
Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
6 / 23 (G, H)-Isomorphism Game
Intuition: Alice and Bob want to convince a referee that G =∼ H.
R To win players must reply h, h0 g g0 • such that rel(h, h0) = rel(g, g0)
No communication during game • A B h h0
Fact. G =∼ H Classical players can win the game with certainty ⇔ Def. (Quantum isomorphism) ∼ 1 We say that G =qc H if quantum players can win the game with certainty.
1We work in the commuting rather than the tensor-product model. 6 / 23 Eg = {Egh B(H): h V(H)} where ∈ ∈ Egh 0, Egh = I. h P
ψ is a unit vector in a Hilbert space H Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈
Bob measures with F 0 • g E and F 0 0 commute • gh g h
Alice and Bob share a quantum state ψ •
The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
R g g0
A h h0 B
ψ
7 / 23 ψ is a unit vector in a Hilbert space H
Eg = {Egh B(H): h V(H)} where ∈ ∈ Egh 0, Egh = I. h P
E and F 0 0 commute • gh g h
The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ R • g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A h h0 B
ψ
Bob measures with F 0 • g
7 / 23 Eg = {Egh B(H): h V(H)} where ∈ ∈ Egh 0, Egh = I. h P
E and F 0 0 commute • gh g h
The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ • R ψ is a unit vector in a Hilbert space H g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A h h0 B
ψ
Bob measures with F 0 • g
7 / 23 E and F 0 0 commute • gh g h
The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ • R ψ is a unit vector in a Hilbert space H g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A B Eg = {Egh B(H): h V(H)} where h h0 ∈ ∈ Egh 0, h Egh = I. ψ
Bob measures with FP0 • g
7 / 23 The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ • R ψ is a unit vector in a Hilbert space H g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A B Eg = {Egh B(H): h V(H)} where h h0 ∈ ∈ Egh 0, h Egh = I. ψ
Bob measures with FP0 • g E and F 0 0 commute • gh g h
7 / 23 The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ • R ψ is a unit vector in a Hilbert space H g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A B Eg = {Egh B(H): h V(H)} where h h0 ∈ ∈ Egh 0, h Egh = I. ψ
Bob measures with FP0 • g E and F 0 0 commute • gh g h
7 / 23 The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ • R ψ is a unit vector in a Hilbert space H g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A B Eg = {Egh B(H): h V(H)} where h h0 ∈ ∈ Egh 0, h Egh = I. ψ
Bob measures with FP0 • g E and F 0 0 commute • gh g h
7 / 23 Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ • R ψ is a unit vector in a Hilbert space H g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A B Eg = {Egh B(H): h V(H)} where h h0 ∈ ∈ Egh 0, h Egh = I. ψ
Bob measures with FP0 • g E and F 0 0 commute • gh g h
The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
7 / 23 Quantum commuting strategies ∼ G =qc H := Quantum players can win the (G, H)-isomorphism game
Alice and Bob share a quantum state ψ • R ψ is a unit vector in a Hilbert space H g g0 Upon receiving g, Alice performs a local • measurement Eg to get h V(H) ∈ A B Eg = {Egh B(H): h V(H)} where h h0 ∈ ∈ Egh 0, h Egh = I. ψ
Bob measures with FP0 • g E and F 0 0 commute • gh g h
The probability that players respond with h, h0 on questions g, g0 is
p(h, h0|g, g0) = ψ, EghFg h ψ h 0 0 i
7 / 23 Example: G =∼ H but G =∼ H 6 qc
x1 + x2 + x3 = 0 x4 + x5 + x6 = 0 x7 + x8 + x9 = 0
000 011 101 110 000 011 101 110 000 011 101 110
000 011 101 110 000 011 101 110 111 100 010 001
x1 + x4 + x7 = 0 x2 + x5 + x8 = 0 x3 + x6 + x9 = 1
Construction based on reduction from linear system games.
8 / 23 Example: G =∼ H but G =∼ H 6 qc
x1 + x2 + x3 = 0 x4 + x5 + x6 = 0 x7 + x8 + x9 = 0
000 011 101 110 000 011 101 110 000 011 101 110
000 011 101 110 000 011 101 110 000 011 101 110
x1 + x4 + x7 = 0 x2 + x5 + x8 = 0 x3 + x6 + x9 = 0
Construction based on reduction from linear system games.
9 / 23 p is a projection, i.e., p2 = p = p for all i, j • ij ij ij ij∗ p = 1 = p for all i, j • k ik ` `j P P Remark. A QPM with entries from C is a permutation matrix.
Thm. (Lupini, M., Roberson) ∼ G =qc H PAGP† = AH for some quantum ⇔ permutation matrix P
Quantum isomorphism and quantum groups
Def. A matrix P = (pij) whose entries are elements of a C∗-algebra is a quantum permutation matrix (QPM), if
10 / 23 p = 1 = p for all i, j • k ik ` `j P P Remark. A QPM with entries from C is a permutation matrix.
Thm. (Lupini, M., Roberson) ∼ G =qc H PAGP† = AH for some quantum ⇔ permutation matrix P
Quantum isomorphism and quantum groups
Def. A matrix P = (pij) whose entries are elements of a C∗-algebra is a quantum permutation matrix (QPM), if p is a projection, i.e., p2 = p = p for all i, j • ij ij ij ij∗
10 / 23 Remark. A QPM with entries from C is a permutation matrix.
Thm. (Lupini, M., Roberson) ∼ G =qc H PAGP† = AH for some quantum ⇔ permutation matrix P
Quantum isomorphism and quantum groups
Def. A matrix P = (pij) whose entries are elements of a C∗-algebra is a quantum permutation matrix (QPM), if p is a projection, i.e., p2 = p = p for all i, j • ij ij ij ij∗ p = 1 = p for all i, j • k ik ` `j P P
10 / 23 Thm. (Lupini, M., Roberson) ∼ G =qc H PAGP† = AH for some quantum ⇔ permutation matrix P
Quantum isomorphism and quantum groups
Def. A matrix P = (pij) whose entries are elements of a C∗-algebra is a quantum permutation matrix (QPM), if p is a projection, i.e., p2 = p = p for all i, j • ij ij ij ij∗ p = 1 = p for all i, j • k ik ` `j P P Remark. A QPM with entries from C is a permutation matrix.
10 / 23 Quantum isomorphism and quantum groups
Def. A matrix P = (pij) whose entries are elements of a C∗-algebra is a quantum permutation matrix (QPM), if p is a projection, i.e., p2 = p = p for all i, j • ij ij ij ij∗ p = 1 = p for all i, j • k ik ` `j P P Remark. A QPM with entries from C is a permutation matrix.
Thm. (Lupini, M., Roberson) ∼ G =qc H PAGP† = AH for some quantum ⇔ permutation matrix P
10 / 23 Can we describe quantum isomorphism in combinatorial terms?
11 / 23 Example
C7 C5 →
hom(F, G) := # of homomorphisms from F to G.
Graph homomorphisms
Def. A map ϕ : V(F) V(G) is a homomorphism from F to G if → ϕ(u) ∼ ϕ(v) whenever u ∼ v.
12 / 23 hom(F, G) := # of homomorphisms from F to G.
Graph homomorphisms
Def. A map ϕ : V(F) V(G) is a homomorphism from F to G if → ϕ(u) ∼ ϕ(v) whenever u ∼ v.
Example
C7 C5 →
12 / 23 Graph homomorphisms
Def. A map ϕ : V(F) V(G) is a homomorphism from F to G if → ϕ(u) ∼ ϕ(v) whenever u ∼ v.
Example
C7 C5 →
hom(F, G) := # of homomorphisms from F to G.
12 / 23 Theorem. (Lov´asz,1967) Homomorphism counts determine a graph up to isomorphism, i.e.
G =∼ H hom(F, G) = hom(F, H) for all graphs F. ⇔
Theorem. (M., Roberson) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F. ⇔
Counting homomorphisms
13 / 23 Theorem. (M., Roberson) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F. ⇔
Counting homomorphisms
Theorem. (Lov´asz,1967) Homomorphism counts determine a graph up to isomorphism, i.e.
G =∼ H hom(F, G) = hom(F, H) for all graphs F. ⇔
13 / 23 Counting homomorphisms
Theorem. (Lov´asz,1967) Homomorphism counts determine a graph up to isomorphism, i.e.
G =∼ H hom(F, G) = hom(F, H) for all graphs F. ⇔
Theorem. (M., Roberson) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F. ⇔
13 / 23 Folklore. G and H cospectral hom(F, G) = hom(F, H) for all cycles F ⇔ Thm. (Dvoˇr´ak,2010; Dell, Grohe, Rattan, 2018) ∼ G =f H hom(F, G) = hom(F, H) for all trees F ∼ ⇔ G =k H hom(F, G) = hom(F, H) for all F of treewidth k ⇔ 6
Complexity: Except for the class of planar graphs, equality of homomorphism counts from all of the above graph classes can be tested in at worst quasi-polynomial time.
Context: Homomorphism counting Thm. (Lov´asz,1967) G =∼ H hom(F, G) = hom(F, H) for all graphs F ⇔ Thm. (M., Roberson, 2019) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F ⇔
14 / 23 Thm. (Dvoˇr´ak,2010; Dell, Grohe, Rattan, 2018) ∼ G =f H hom(F, G) = hom(F, H) for all trees F ∼ ⇔ G =k H hom(F, G) = hom(F, H) for all F of treewidth k ⇔ 6
Complexity: Except for the class of planar graphs, equality of homomorphism counts from all of the above graph classes can be tested in at worst quasi-polynomial time.
Context: Homomorphism counting Thm. (Lov´asz,1967) G =∼ H hom(F, G) = hom(F, H) for all graphs F ⇔ Thm. (M., Roberson, 2019) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F ⇔ Folklore. G and H cospectral hom(F, G) = hom(F, H) for all cycles F ⇔
14 / 23 ∼ G =k H hom(F, G) = hom(F, H) for all F of treewidth k ⇔ 6
Complexity: Except for the class of planar graphs, equality of homomorphism counts from all of the above graph classes can be tested in at worst quasi-polynomial time.
Context: Homomorphism counting Thm. (Lov´asz,1967) G =∼ H hom(F, G) = hom(F, H) for all graphs F ⇔ Thm. (M., Roberson, 2019) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F ⇔ Folklore. G and H cospectral hom(F, G) = hom(F, H) for all cycles F ⇔ Thm. (Dvoˇr´ak,2010; Dell, Grohe, Rattan, 2018) ∼ G =f H hom(F, G) = hom(F, H) for all trees F ⇔
14 / 23 Complexity: Except for the class of planar graphs, equality of homomorphism counts from all of the above graph classes can be tested in at worst quasi-polynomial time.
Context: Homomorphism counting Thm. (Lov´asz,1967) G =∼ H hom(F, G) = hom(F, H) for all graphs F ⇔ Thm. (M., Roberson, 2019) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F ⇔ Folklore. G and H cospectral hom(F, G) = hom(F, H) for all cycles F ⇔ Thm. (Dvoˇr´ak,2010; Dell, Grohe, Rattan, 2018) ∼ G =f H hom(F, G) = hom(F, H) for all trees F ∼ ⇔ G =k H hom(F, G) = hom(F, H) for all F of treewidth k ⇔ 6
14 / 23 : Except for the class of planar graphs, equality of homomorphism counts from all of the above graph classes can be tested in at worst quasi-polynomial time.
Context: Homomorphism counting Thm. (Lov´asz,1967) G =∼ H hom(F, G) = hom(F, H) for all graphs F ⇔ Thm. (M., Roberson, 2019) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F ⇔ Folklore. G and H cospectral hom(F, G) = hom(F, H) for all cycles F ⇔ Thm. (Dvoˇr´ak,2010; Dell, Grohe, Rattan, 2018) ∼ G =f H hom(F, G) = hom(F, H) for all trees F ∼ ⇔ G =k H hom(F, G) = hom(F, H) for all F of treewidth k ⇔ 6
Complexity
14 / 23 Context: Homomorphism counting Thm. (Lov´asz,1967) G =∼ H hom(F, G) = hom(F, H) for all graphs F ⇔ Thm. (M., Roberson, 2019) ∼ G =qc H hom(F, G) = hom(F, H) for all planar graphs F ⇔ Folklore. G and H cospectral hom(F, G) = hom(F, H) for all cycles F ⇔ Thm. (Dvoˇr´ak,2010; Dell, Grohe, Rattan, 2018) ∼ G =f H hom(F, G) = hom(F, H) for all trees F ∼ ⇔ G =k H hom(F, G) = hom(F, H) for all F of treewidth k ⇔ 6
Complexity: Except for the class of planar graphs, equality of homomorphism counts from all of the above graph classes can be tested in at worst quasi-polynomial time.
14 / 23 Before: Difficult to prove that they are not quantum isomorphic.
Now: Only one (the Rook graph) contains K4.
Application: Certificate for G =∼ H 6 qc Are these two graphs quantum isomorphic? Rook graph Shrikhande graph
15 / 23 Now: Only one (the Rook graph) contains K4.
Application: Certificate for G =∼ H 6 qc Are these two graphs quantum isomorphic? Rook graph Shrikhande graph
Before: Difficult to prove that they are not quantum isomorphic.
15 / 23 Application: Certificate for G =∼ H 6 qc Are these two graphs quantum isomorphic? Rook graph Shrikhande graph
Before: Difficult to prove that they are not quantum isomorphic.
Now: Only one (the Rook graph) contains K4. 15 / 23 Part II Elements of the proof
∼ Thm. G =qc H hom(F, G) = hom(F, H) for all planar graphs F ⇔ Main component of our proof: Provide a combinatorial description of the intertwiners of Qut(G).
Intertwiners of Qut(G) = U, M, AG , , ,lin, where h i◦ ⊗ ∗
U = ei, M(ei ej) = δijei i, j V(G). ⊗ ∀ ∈ i V(G) ∈X
The starting point
Thm. (Lupini, M., Roberson) Quantum isomorphism can be characterized in terms of the quantum automorphism group of a graph, denoted Qut(G).
17 / 23 Intertwiners of Qut(G) = U, M, AG , , ,lin, where h i◦ ⊗ ∗
U = ei, M(ei ej) = δijei i, j V(G). ⊗ ∀ ∈ i V(G) ∈X
The starting point
Thm. (Lupini, M., Roberson) Quantum isomorphism can be characterized in terms of the quantum automorphism group of a graph, denoted Qut(G).
Main component of our proof: Provide a combinatorial description of the intertwiners of Qut(G).
17 / 23 The starting point
Thm. (Lupini, M., Roberson) Quantum isomorphism can be characterized in terms of the quantum automorphism group of a graph, denoted Qut(G).
Main component of our proof: Provide a combinatorial description of the intertwiners of Qut(G).
Intertwiners of Qut(G) = U, M, AG , , ,lin, where h i◦ ⊗ ∗
U = ei, M(ei ej) = δijei i, j V(G). ⊗ ∀ ∈ i V(G) ∈X
17 / 23 Example. ~F = K4, (2, 1), (2, 2)