C*-algebras and the C*-identity
Connect the adjoint, multiplication, and norm through the C*-identity and test why seemingly reasonable matrix norms can fail it.
On this page
The question for today
What you’ll understand
- Verify the conjugate-linear order-reversing involution rules.
- Use the C*-identity to connect a matrix norm with a positive product.
- Distinguish a complete C*-algebra from a dense nonclosed *-subalgebra.
The adjoint is not an extra decoration on a matrix. Together with the norm, it encodes a positivity structure that survives beyond finite-dimensional coordinates. A complex C*-algebra is a Banach algebra with an involution and a particular norm identity. We initially assume a unit to keep the formulas short, then explicitly discuss a nonunital example. The distinction between an abstract algebra and a chosen operator representation will become important when constructing states and representations.
The involution reverses products and conjugates scalars
An involution sends a to a-star, is conjugate-linear, reverses multiplication order, and squares to the identity operation. In complex matrices it is conjugate transpose, not plain transpose. In continuous complex-valued functions it is pointwise complex conjugation. An element is self-adjoint when it equals its adjoint, normal when it commutes with its adjoint, and unitary when multiplying by its adjoint on either side gives the unit. These are different conditions.
The C*-identity says the norm of a-star times a equals the square of the norm of a. Along with the Banach-algebra structure, it makes the involution isometric. For bounded operators on a complex Hilbert space, the identity follows from the definition of the adjoint and the operator norm. Using another matrix norm can preserve submultiplicativity yet destroy this identity, so calling an algebra a C*-algebra requires checking the compatible norm rather than only listing its operations.
Examples reveal why completeness and the unit matter
The algebra B(H) of all bounded operators on a complex Hilbert space is unital and complete in operator norm. A norm-closed *-subalgebra is again a C*-algebra, though it may not contain B(H)'s identity. On an infinite-dimensional Hilbert space, the compact operators form a nonunital C*-algebra. The finite-rank operators form a *-algebra but are not norm closed; their completion is the compact-operator algebra. Thus finite-dimensional approximability does not itself supply a unit or completeness.
For a locally compact Hausdorff space X, continuous functions vanishing at infinity form C₀(X), with supremum norm and pointwise operations. If X is compact, this becomes C(X), containing the constant function one. If X is noncompact, that constant does not vanish at infinity. This provides a commutative nonunital example without using matrices. Whenever we later write one or normalize a state by its value at one, we will explicitly be in the unital setting or use the appropriate nonunital definition.
The algebraic *-structure tightly controls its norm
For a normal element of a C*-algebra, the spectral radius equals its norm. Applying this to the positive element a-star a links the norm to spectral information intrinsic to the *-algebra. In particular, there is at most one C*-norm making a fixed unital *-algebra complete. This rigidity is stronger than the flexibility of norms on a general vector space. It does not say that every abstract *-algebra admits such a completion or such a norm.
A *-homomorphism between C*-algebras is automatically contractive, and an injective one is isometric. In the unital case, one proof first compares spectra of self-adjoint elements and then uses the C*-identity for general elements. These conclusions concern maps preserving the adjoint; an arbitrary algebraic linear map need not share them. The representation theorem that every C*-algebra admits a faithful Hilbert-space representation will be reached through states, rather than assumed as the definition of every abstract algebra.
Work it through
Test two norms on the same matrix algebra
Let D=diag(1,2) in M₂(C). Verify the C*-identity with operator norm, then show that the Frobenius norm fails it already at I₂.
The operator norm of D is 2, and D*D=diag(1,4) has operator norm 4. Therefore ||D*D||=4=||D||².
The Frobenius norm is the square root of the sum of squared entry moduli. Hence ||I₂||F=√2 and ||I₂*I₂||F=√2.
The required square would be ||I₂||F²=2, which differs from √2. Thus this norm does not make the usual matrix *-algebra a C*-algebra, despite being a useful complete matrix norm.
The operation alone is insufficient: operator norm satisfies the C*-identity, while Frobenius norm with the same product and adjoint does not.
The living notebook
Explore the model
Interactive lab · matrix
Compare a matrix with its singular values
Pause, reset, then set diagonal scale a=2 and shear b=0. Predict singular values 2 and 1, determinant 2, and condition number 2. The operator norm is therefore 2 and the norm of A*A is 4. The instrument displays the selected triangular matrix and its singular values; it does not let you switch to Frobenius norm or represent all C*-algebras.
- Area scale · det A
- 1.4
- Singular values · σmax, σmin
- 1.597, 0.876
- Condition number · κ₂(A)
- 1.823
The selected matrix is A = [[a,b],[0,1]]. Playback interpolates geometry between the identity and A; readouts always describe A. Its determinant is a and its singular values come from AᵀA. Positive a makes every supported matrix invertible. This two-dimensional family is not a general eigensolver or a learned network. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forThe C*-identity couples the operator norm to the adjoint product, rather than to an arbitrary entrywise matrix norm.
Pause & reason
Check your understanding
Your turn
Try it yourself
For A=[[0,3],[0,0]], compute A*A, AA*, the operator norm of A, and whether A is normal.
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
Take the conjugate transpose before multiplying. The norm follows from the lengths of possible outputs.
Compare with the solution
A*=[[0,0],[3,0]], so A*A=diag(0,9) and AA*=diag(9,0). They differ, hence A is not normal. Acting on (x,y) gives (3y,0), so the operator norm is 3. The norm of A*A is 9, agreeing with the C*-identity even though A itself is nonnormal. Normality is not a hypothesis for that identity.
Keep the idea
Take it with you
- The involution is conjugate-linear and reverses multiplication order.
- The C*-identity singles out a norm compatible with multiplication and adjoints.
- Norm closure and presence of a unit are separate structural questions.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.