Generalize infsiteinds to fractional QN densities

src/infinitecanonicalmps.jl

@@ -5,18 +5,47 @@ end
 
 # TODO: Move to ITensors.jl
 function Base.:/(qnval::ITensors.QNVal, n::Int)
-  div_val = ITensors.val(qnval) / n
-  if !isinteger(div_val)
-    error("Dividing $qnval by $n, the resulting QN value is not an integer")
+  if abs(qnval.modulus) <= 1
+    div_val = ITensors.val(qnval) / n
+    if !isinteger(div_val)
+      error("Dividing $qnval by $n, the resulting QN value is not an integer")
+    end
+    return setval(qnval, Int(div_val))
+  else
+    if mod(qnval.val, gcd(qnval.modulus, n)) != 0
+      error("Dividing $qnval by $n, no solution to the Chinese remainder theorem")
+    end
+    #We perform a brute force sieving solution -> should be ok for all reasonnable n
+    sol = qnval.val
+    for x in 1:n
+      if mod(sol, n) != 0
+        sol += qnval.modulus
+      else
+        break
+      end
+      x == n && error("Failed to find solution") #Bezut identity would be a better solver
+    end
+    modulus = lcm(qnval.modulus, n)
+    sol = mod(sol, modulus)
+    sol = sol < abs(sol - modulus) ? sol : sol - modulus
+    return setval(qnval, Int(sol / n))
   end
-  return setval(qnval, Int(div_val))
+end
+
+function Base.:*(qnval::ITensors.QNVal, n::Int)
+  prod_val = ITensors.val(qnval) * n
+  return setval(qnval, Int(prod_val))
 end
 
 # TODO: Move to ITensors.jl
 function Base.:/(qn::QN, n::Int)
   return QN(map(qnval -> qnval / n, qn.data))
 end
 
+function Base.:*(qn::QN, n::Int)
+  return QN(map(qnval -> qnval * n, qn.data))
+end
+
 # of Index (Tuple, Vector, ITensor, etc.)
 indtype(i::Index) = typeof(i)
 indtype(T::Type{<:Index}) = T
@@ -48,11 +77,69 @@ function shift_flux(i::Index, flux_density::QN)
   return ITensors.setspace(i, shift_flux(space(i), flux_density))
 end
 
+function scale_flux(qn::ITensors.QNVal, flux_factor::Int)
+  flux_factor == 1 && return qn #This test could occur earlier
+  if -1 <= qn.modulus <= 1  #U(1) symmetry, should remain a U(1) symmetry, if qn.modulus = 0, it is a dummy term and the code does nothing
+    return qn * flux_factor
+  else #To get the same algebra, multiplying the modulus by flux_factor works
+    return ITensors.QNVal(qn.name, qn.val * flux_factor, qn.modulus * flux_factor)
+  end
+end
+
+function scale_flux(qn::ITensors.QNVal, flux_factor::Dict{ITensors.SmallString,Int})
+  !haskey(flux_factor, qn.name) && return qn #Ensure that we do not touch empty QNs or QNs that we do not need to rescale because the shift is 0
+  return scale_flux(qn, flux_factor[qn.name])
+end
+
+function scale_flux(qn::QN, flux_factor::Union{Int,Dict{ITensors.SmallString,Int}})
+  return QN(map(qnval -> scale_flux(qnval, flux_factor), qn.data))
+end
+
+function scale_flux(
+  qnblock::Pair{QN,Int}, flux_factor::Union{Int,Dict{ITensors.SmallString,Int}}
+)
+  return (scale_flux(ITensors.qn(qnblock), flux_factor) => ITensors.blockdim(qnblock))
+end
+
+function scale_flux(
+  space::Vector{Pair{QN,Int}}, flux_factor::Union{Int,Dict{ITensors.SmallString,Int}}
+)
+  return map(qnblock -> scale_flux(qnblock, flux_factor), space)
+end
+
+function scale_flux(i::Index, flux_factor::Union{Int,Dict{ITensors.SmallString,Int}})
+  return ITensors.setspace(i, scale_flux(space(i), flux_factor))
+end
+
 function shift_flux_to_zero(s::Vector{<:Index}, flux::QN)
   if iszero(flux)
     return s
   end
+  #We are introducing a factor per QN
   n = length(s)
+  multipliers = Dict{ITensors.SmallString,Int}() #multipliers is assigned using the names
+  for qn in flux.data
+    if qn.val == 0 #We do not need to multiply in this case. It also takes into account the empty QNs which have val and modulus 0
+      continue
+    end
+    multipliers[qn.name] = abs(lcm(qn.val, n) ÷ qn.val)#default solution with positive multiplier
+    if abs(qn.modulus) > 1
+      #We find a multiplier which guarantees a solution of the Chinese theorem
+      #For now, brute force:
+      multipliers[qn.name] = abs(lcm(qn.val, n) ÷ qn.val)
+      for x in 1:(multipliers[qn.name] - 1)
+        if mod(n, x) != 0
+          continue
+        end
+        if mod(qn.val * x, gcd(n, x * qn.modulus)) == 0
+          multipliers[qn.name] = x
+          break
+        end
+      end
+    end
+  end
+  s = map(sₙ -> scale_flux(sₙ, multipliers), s)
+  flux = scale_flux(flux, multipliers)
   flux_density = flux / n
   return map(sₙ -> shift_flux(sₙ, flux_density), s)
 end

src/subspace_expansion.jl

@@ -18,7 +18,7 @@ function generate_twobody_nullspace(
 
   range_H = nrange(ψ, H[1])
 
-  @assert range_H > 1 "Not defined for purely local Hamiltonians"
+  @assert range_H > 1 "Subspace expansion for InfiniteSum{MPO} is not defined for purely local Hamiltonians"
 
   if range_H > 2
     ψᴴ = dag(ψ)
@@ -181,7 +181,7 @@ function subspace_expansion(
   @assert dˡ == dʳ
   if dˡ ≥ maxdim
     println(
-      "Current bond dimension at bond $b is $dˡ while desired maximum dimension is $maxdim, skipping bond dimension increase",
+      "Current bond dimension at bond $b is $dˡ while desired maximum dimension is $maxdim, skipping bond dimension increase at $b",
     )
     flush(stdout)
     flush(stderr)
@@ -199,7 +199,7 @@ function subspace_expansion(
   #Added due to crash during testing
   if norm(ψHN2.tensor) < 1e-12
     println(
-      "Impossible to do a subspace expansion, probably due to conservation constraints"
+      "The two-site subspace expansion produced a zero-norm expansion at $b. This is likely due to the long-range nature of the QN conserving Hamiltonian.",
     )
     flush(stdout)
     flush(stderr)

test/test_unitcell_qns.jl

@@ -0,0 +1,47 @@
+using ITensors
+using ITensorInfiniteMPS
+using Test
+using Random
+
+@testset "unitcell_qns" begin
+  function initstate(n)
+    if n % 2 == 0
+      if (n / 2) % 2 == 1
+        return "Dn"
+      else
+        return "Up"
+      end
+    else
+      if (n + 1) / 2 <= 19
+        if ((n + 1) / 2) % 2 == 1
+          return "Up"
+        else
+          return "Dn"
+        end
+      else
+        return "Emp"
+      end
+    end
+  end
+  conserve_qns = true
+  N = 62
+  s = infsiteinds(n -> isodd(n) ? "tJ" : "S=1/2", N; conserve_qns, initstate)
+  ψ = InfMPS(s, initstate)
+  @show N
+  @show s
+
+  function initstate(n)
+    if mod(n, 4) == 1
+      return 2
+    else
+      return 1
+    end
+  end
+
+  for N in 1:6
+    s = infsiteinds("S=1/2", N; conserve_szparity=true, initstate)
+    ψ = InfMPS(s, initstate)
+    @show N
+    @show s
+  end
+end

test/test_vumpsmpo.jl

@@ -9,7 +9,7 @@ function expect_three_site(ψ::MPS, h::ITensor, n::Int)
 end
 
 #Ref time is 21.6s with negligible compilation time
-@time @testset "vumpsmpo_ising" begin
+@testset "vumpsmpo_ising" begin
   Random.seed!(1234)
 
   model = Model("ising")

test/test_vumpsmpo_fqhe.jl

@@ -21,7 +21,7 @@ end
 
 #Currently, VUMPS cannot give the right result as the subspace expansion is too small
 #This is meant to test the generalized translation rules
-@time @testset "vumpsmpo_fqhe" begin
+@testset "vumpsmpo_fqhe" begin
   Random.seed!(1234)
   function initstate(n)
     if mod(n, 3) == 1